Dynamic calibration method and system for empty body weight of giant panda based on multi-modal data fusion

Abstract

This invention provides a method and system for dynamic calibration of fasting weight of giant pandas using multimodal data fusion, relating to the field of wildlife conservation and breeding research technology. The method includes first acquiring multimodal data of giant pandas in their natural state; then constructing an environment-morphology-motion-centroid joint dynamic embedding set through environment-context-aware Koopman-centroid dynamics joint embedding processing; next, using persistent cohomology and knot theory for topological filtering to extract a centroid drift topological feature sequence robust to environmental fluctuations; further, obtaining a centroid-energy joint state probability measure through environment-inertia weighted hypergraph fusion and joint state symbolization processing; and finally, obtaining the dynamic calibration result of the giant panda's fasting weight through joint compensation of centroid drift and environmental fluctuations. This invention effectively overcomes the limitations of traditional contact-based measurements, achieving high-precision dynamic monitoring of fasting weight under non-interference conditions.

Description

Technical Field

[0001] This invention relates to the field of wildlife conservation and breeding research technology, and more specifically, to a method and system for dynamic calibration of fasting weight of giant pandas using multimodal data fusion. Background Technology

[0002] Giant pandas are rare and protected animals in my country, known as "living fossils" and "China's national treasure," and are a flagship species for global biodiversity conservation. In the field of giant panda conservation and breeding research, accurately obtaining the fasting weight of giant pandas is crucial for assessing their health status, nutritional level, and reproductive potential.

[0003] Currently, in the ex-situ conservation and scientific research of giant pandas, the weight monitoring of captive giant pandas mainly relies on traditional weighing scales. Weighing scales requires luring individuals to stand on the platform, which has problems such as interfering with natural behavior, stress reactions affecting the accuracy of measurements, and the inability to achieve high-frequency continuous monitoring.

[0004] Weighing pandas in a non-fasting state presents a fundamental challenge: after eating, the presence of food and gas in the gastrointestinal tract increases their mass, causing the weighing results to deviate significantly from their true fasting weight. Furthermore, microclimate fluctuations (temperature, humidity, airflow) in wild or semi-natural captive environments significantly modulate the distribution of surface thermal infrared radiation and fur fluffiness, further interfering with the analysis of fasting weight and reducing the reliability of the results. Current technologies cannot fundamentally eliminate the systematic biases caused by these dynamic physiological and environmental interference sources, resulting in large fluctuations and poor stability in measurement results, and failing to accurately reflect the animal's true physiological state in its natural state.

[0005] Therefore, there is an urgent need for a method and system for dynamic calibration of fasting weight of giant pandas using multimodal data fusion to solve the above-mentioned technical problems. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for dynamic calibration of fasting weight of giant pandas using multimodal data fusion, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0007] Firstly, this application provides a method for dynamic calibration of fasting weight of giant pandas using multimodal data fusion, including:

[0008] Acquire multimodal data of giant pandas in their natural state. The multimodal data includes three-dimensional body surface depth image data at multiple time points, gait acceleration sequence, body surface thermal infrared distribution time series, event timestamp sequence of feeding behavior, metabolic marker time series before and after each feeding, trunk inertial measurement data, and spatiotemporal sequence of local environmental climate field.

[0009] Based on the multimodal basic data, a context-aware Koopman-centroid dynamics joint embedding process is performed to obtain an environment-morphology-motion-centroid joint dynamics embedding set;

[0010] Based on the environment-morphology-motion-centroid joint dynamic embedding set, filtering based on persistent cohomology and knot theory is performed to obtain the centroid drift topological feature sequence.

[0011] Based on the centroid drift topological feature sequence and the multimodal data, a discretized centroid-energy joint state probability measure is obtained by performing environment-inertia weighted hypergraph fusion and joint state symbolization processing.

[0012] Based on the discretized centroid-energy joint state probability measure, a quantum dual-interference compensation-based inversion process was performed to obtain the dynamic calibration results of the fasting weight of giant pandas, which compensated for centroid drift and environmental fluctuation effects.

[0013] Secondly, this application also provides a multimodal data fusion-based dynamic calibration system for fasting weight of giant pandas, comprising:

[0014] The acquisition unit is used to acquire multimodal data of giant pandas in their natural state. The multimodal data includes three-dimensional body surface depth image data at multiple times, gait acceleration sequence, body surface thermal infrared distribution time series, event timestamp sequence of feeding behavior, metabolic marker time series before and after each feeding, trunk inertial measurement data, and spatiotemporal sequence of local environmental climate field.

[0015] The processing unit is used to perform environment context-aware Koopman-centroid dynamics joint embedding processing based on the multimodal basic data to obtain an environment-morphology-motion-centroid joint dynamics embedding set.

[0016] The filtering unit is used to perform filtering based on persistent cohomology and knot theory according to the environment-morphology-motion-centroid joint dynamic embedding set to obtain the centroid drift topological feature sequence.

[0017] The fusion unit is used to perform environment-inertia weighted hypergraph fusion and joint state symbolization processing based on the centroid drift topological feature sequence and the multimodal data to obtain a discretized centroid-energy joint state probability measure.

[0018] The inversion unit is used to perform quantum dual-interference compensation inversion processing based on the discretized centroid-energy joint state probability measure to obtain the dynamic calibration result of the fasting weight of giant pandas that has been compensated for centroid drift and environmental fluctuation effects.

[0019] The beneficial effects of this invention are as follows:

[0020] This invention revolutionizes weight measurement from a static "mass estimation" to a dynamic "mass-center-of-mass joint estimation" by deeply fusing multi-source heterogeneous data, including 3D body surface depth images, gait acceleration, body surface thermal infrared data, metabolic markers, behavioral events, and environmental microclimate. First, it introduces a centroid dynamic model based on Newton-Euler equations and inertial measurement unit data to explicitly characterize the centroid drift effect caused by internal mass migration during fasting. Then, it linearizes the nonlinear dynamic system using Koopman operator theory and innovatively combines persistent cohomology and knot theory to extract robust centroid drift topological features from high-dimensional dynamic embeddings. Based on this, it constructs an environment-inertial weighted hypergraph to fuse the centroid, thermometabolism, and environmental modes, symbolizing them as a "centroid-energy" joint state. Finally, it performs joint inversion and compensation for two types of disturbances: centroid drift and environmental fluctuations, outputting a dynamic weight calibration value that is time-continuous, topologically robust, and accurately reflects the physiological fasting state.

[0021] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the process for dynamic calibration of fasting weight of giant pandas using multimodal data fusion as described in this embodiment of the invention.

[0024] Figure 2 This is a schematic diagram of the structure of the multimodal data fusion dynamic calibration system for fasting weight of giant pandas described in this embodiment of the invention.

[0025] In the diagram: 701, acquisition unit; 702, processing unit; 703, filtering unit; 704, fusion unit; 705, inversion unit. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0028] Example 1:

[0029] See Figure 1 This embodiment provides a method for dynamic calibration of fasting weight of giant pandas using multimodal data fusion, including steps S1, S2, S3, S4 and S5.

[0030] Step S1: Obtain multimodal data of giant pandas in their natural state. The multimodal data includes three-dimensional body surface depth image data at multiple times, gait acceleration sequence, body surface thermal infrared distribution time series, event timestamp sequence of feeding behavior, metabolic marker time series before and after each feeding, trunk inertial measurement data, and spatiotemporal sequence of local environmental climate field.

[0031] It is understandable that the multiple time points in this step refer to all moments within 30 minutes before and after the giant panda eats, and the natural state refers to the physiological and behavioral state of the giant panda while maintaining its inherent behavioral patterns. A depth camera was used to collect a sequence of surface point clouds with three-dimensional spatial coordinates and temporal relationships, which depicts the dynamic changes in morphological features such as the abdominal contour (three-dimensional surface depth image data); a high-speed camera captured video of the giant panda's movement, and its gait acceleration sequence was determined based on the displacement changes within the video; an inertial measurement unit (IMU) sensor attached to the trunk recorded the acceleration and angular velocity sequence of the overall movement, reflecting the inertial dynamics of the movement posture and the trunk itself (trunk inertial measurement data); an infrared thermal imager captured the temporal sequence of surface temperature distribution, characterizing local metabolic heat production and heat dissipation (surface thermal infrared distribution temporal sequence); and behavioral observation logs were used to record... The study collected precise feeding event timestamps (timestamp sequences of feeding behavior events) and short-term metabolic rate surrogate indicators (time series of metabolic markers before and after each feeding) indirectly obtained through metabolic measurement devices. The short-term metabolic rate was calculated by weighing excrement and dividing its weight by body weight. IMU data provided the only mechanical observational basis for estimating the center-of-mass drift caused by changes in the distribution of internal mass (such as gastrointestinal contents). Meanwhile, the spatiotemporal sequences of temperature, humidity, and airflow velocity (spatiotemporal sequences of the local environmental climate field) collected by a microclimate sensor network deployed within the activity area constructed a quantitative field model of environmental microclimate fluctuations, used to decouple the modulation effects of the environment on body surface thermal signals and apparent morphology. This step not only collected conventional physiological parameters but also innovatively established the IMU kinetic signals and microclimate field environmental signals as core foundational data. This provides an irreplaceable data foundation for subsequent modeling and compensation of center-of-mass drift and environmental interference. The technical effect is the generation of a multi-dimensional, spatiotemporally aligned dataset containing quantitative information on specific interference factors, providing high-quality and physically meaningful input for the entire calibration process.

[0032] Step S2: Perform environment context-aware Koopman-centroid dynamics joint embedding processing based on the multimodal basic data to obtain the environment-morphology-motion-centroid joint dynamics embedding set;

[0033] Understandably, this step treats the giant panda's torso as a dynamic system. It uses the acceleration and angular velocity measured by the torso's inertia to infer the net external force and net torque acting on the torso. Then, combining this with the geometric segments obtained from the discretization of the three-dimensional point cloud and their anatomical prior masses, it estimates the dynamic drift of the center of mass caused by the distribution and changes in internal mass (such as gastrointestinal contents). In this step, step S2 includes steps S21, S22, S23, and S24.

[0034] Step S21: Based on the spatiotemporal sequence of the local environmental climate field, environmental driving factors are extracted. Principal component analysis is used to reduce the dimensionality and extract the principal component sequence that changes over time. Then, mutual information analysis is performed between the principal component sequence and the behavioral event timestamp sequence. Environmental principal components with mutual information greater than a preset threshold with eating and activity behaviors are selected as environmental context driving signals.

[0035] Understandably, this step first performs principal component analysis (PCA) to reduce the dimensionality of the spatiotemporal sequence of the local environmental microclimate field (temperature, humidity, airflow, etc.). PCA not only compresses the data dimensionality but, more importantly, extracts the main patterns of coordinated change in the environmental field (such as combined patterns like "synchronous increase in humidity and heat" or "enhanced dry and cold airflow"). The principal component sequences represent the most significant and representative temporal dynamics of environmental fluctuations, thus filtering out minor and irrelevant random fluctuations. Subsequently, this step introduces mutual information analysis for feature selection. Instead of simply selecting the principal component with the largest variance, it calculates the mutual information between each environmental principal component sequence and the event timestamp sequence of feeding behavior. Mutual information can capture any statistical dependence between variables, including nonlinear relationships. The formula for mutual information analysis is shown below:

[0036] ;

[0037] Where I(X;Y) represents the mutual information between the environmental principal component sequence X and the behavioral event sequence Y, p(x,y) represents the joint probability distribution of the environmental principal component sequence X and the behavioral event sequence Y, p(x)p(y) represents the joint marginal probability distribution of the environmental principal component sequence X and the behavioral event sequence Y, x represents the data in the environmental principal component sequence, and y represents the data in the behavioral event sequence.

[0038] This step, by setting the mutual information threshold to 0.2, can automatically filter out those environmental principal components that are statistically significantly related to behavior. For example, it may be found that a particular temperature-humidity combination pattern (characterized by a principal component) is highly correlated with the giant panda's resting behavior after eating.

[0039] Understandably, this step generates a low-dimensional, redundancy-free environmental context-driven signal that is closely tied to the individual's behavioral and physiological state. This signal can more accurately reflect the environmental changes that truly modulate animal physiology, providing high-quality input for constructing a more physiologically reliable extended phase space modulated by the environment.

[0040] Step S22: Based on the three-dimensional body surface depth image data and the trunk inertial measurement unit data, perform the centroid dynamic system construction process. By treating the three-dimensional body surface point cloud of continuous frames as a time-varying massless shell, the net external force and net torque acting on the shell are estimated by inverting the Newton-Euler equation using the trunk inertial measurement unit data. The shell is then discretized into multiple rigid segments and given an equivalent mass based on anatomical priors to obtain a centroid dynamic model that reflects the internal mass transfer.

[0041] Understandably, this step first treats the model constructed from continuous frames of 3D body surface depth image data as a time-varying geometric shell, making the crucial assumption that the shell itself has no mass. The core of the processing lies in utilizing IMU data fixed to the torso, which measures the absolute kinematic information (acceleration and angular velocity) at a point on the shell. Dynamic inversion is performed using the Newton-Euler equations, that is, the cause of the measured motion (acceleration / angular velocity) is deduced (force and torque), thereby estimating the net external force and net torque acting on the shell. This net external force / torque actually includes the effects of internal mass (such as gastrointestinal contents and organs) on the torso shell due to gravity, inertia, etc.

[0042] Next, to distribute the estimated total force across specific mass distributions, the process discretizes the continuous torso shell into multiple anatomically significant rigid segments (e.g., head, thorax, abdomen, and limbs). The crucial step is assigning these segments equivalent masses based on anatomical priors (e.g., deriving the basic tissue mass of each segment from historical anatomical data or population averages). Now, knowing the geometric positions (from point clouds) and equivalent masses of each segment, and combining this with the inverted total external force / moment, a dynamic equation for the entire system (shell + internal mass) can be constructed. By solving this equation, the current state of the internal mass distribution can be deduced, thus dynamically determining the instantaneous center of mass position of the entire torso system.

[0043] The centroid dynamic model includes translational equations, rotational equations, centroid position calculation equations, and constraint equations. The translational equations are shown below:

[0044] ;

[0045] The rotation equations are as follows:

[0046] ;

[0047] The equation for calculating the position of the centroid is shown below:

[0048] ;

[0049] The constraint equations are as follows:

[0050] Where τ is the resultant external force obtained from IMU data inversion, and F is the resultant torque obtained from IMU data inversion. Let Δm be the anatomical prior quality of the i-th segment. i Let a be the change in mass distribution to be solved. i Let g be the acceleration of the i-th segment, g be the gravitational acceleration vector, N be the total number of segments, and I be the acceleration of the i-th segment. i Let be the inertial tensor of the i-th segment about its own center of mass, α be the angular acceleration of the i-th segment, ω be the angular velocity of the i-th segment, R be the instantaneous position of the center of mass, and r be the inertial tensor of the i-th segment about its own center of mass. i Let be the position of the geometric centroid of the i-th segment.

[0051] Step S23: Based on the environmental context driving signal and the centroid dynamics model, construct the extended phase space of the environment-centroid coupling. By introducing the environmental driving signal as an external forcing term into the state equation of the centroid dynamics model, and constructing the extended phase space of the environment-centroid coupling relationship based on the state equation and the gait acceleration sequence.

[0052] Understandably, this step is based on the center-of-mass dynamics model (whose state variables include center-of-mass coordinates and acceleration), and directly introduces the environmental context driving signal from step S21 as an external forcing term into the center-of-mass dynamics model. This means that the center-of-mass dynamics model adds a function related to the environmental signal, thus mathematically establishing a formal driving relationship between the environment and the system dynamics. For example, an increase in environmental temperature may act as an input term, affecting blood flow distribution and muscle tension in the trunk, thereby subtly altering its dynamic characteristics. Subsequently, the processing does not stop at this enhanced environment-system model itself, but uses it as the core, combined with observational data such as gait acceleration sequences, to reconstruct the phase space. It uses the state variables (center-of-mass coordinates, acceleration) of the continuous-time dynamics system, the historical values ​​of the environmental driving signal (historical values ​​of all moments before and after the current driving signal within 30 minutes before and after eating), and the key observational variable (gait acceleration) as coordinate axes to form a higher-dimensional extended phase space. In this space, any instantaneous state of the system is represented by a point, and its change over time forms a trajectory.

[0053] The external forcing term (a function related to environmental signals) is shown below:

[0054] ;

[0055] Where h(E(t),D,t) represents the external forcing term, E(t) is the environmental context driving signal vector, t is the time point, D is the set of the centroid position vector and acceleration vector, and M... -1F e (E(t)) is the driving term of the rate of change of velocity, representing the acceleration caused by the environment, F e M represents the resultant force generated by the environmental signal. -1 I represents the inverse of the diagonal matrix representing mass. -1 τ e (E(t),D) is the driving term of the rate of change of acceleration, I -1 This is the inverse of the moment of inertia matrix, which is composed of angular accelerations, τ. e (E(t),D) represents the resultant torque generated by the environmental signal.

[0056] This step constructs an extended phase space that can simultaneously characterize the giant panda's center-of-mass migration and the driving forces of the external environment. This phase space fully contains the key information of the system's evolution, providing an ideal operational domain with clear physical and physiological significance for the next step of applying Koopman operator theory for global linearization.

[0057] Step S24: Based on the extended phase space, apply the dynamic mode decomposition algorithm to solve the approximate matrix of its Koopman operator, and perform spectral decomposition on the matrix. Select the eigenvector corresponding to the slow-varying eigenmode most relevant to the fasting process, and project the three-dimensional body surface depth image data and gait acceleration sequence at multiple times onto the subspace spanned by these eigenvectors to obtain the final environment-morphology-motion-centroid joint dynamic embedding set.

[0058] Understandably, this step first takes the trajectory data (i.e., the sequence of system state changes over time) in the phase space as input, based on the extended phase space, and applies the Dynamic Mode Decomposition (DMD) algorithm. The DMD algorithm minimizes the prediction error by finding an optimal linear operator (i.e., a finite-dimensional approximation matrix of the Koopman operator) that can describe the linear evolution of the state vector in the phase space from the current time step to the next. This process cleverly transforms the complex nonlinear dynamics problem of the original system into a linear dynamics problem in this high-dimensional embedding space. Subsequently, the Koopman approximation matrix is ​​subjected to spectral decomposition (eigenvalue decomposition) to obtain its eigenvalues ​​(representing the growth rate / decrease rate and oscillation frequency of the modes) and the corresponding eigenvectors (representing the vibration modes in the extended phase space).

[0059] The specific steps are as follows: First, construct a data matrix from the system state change sequence over time. Then, perform Singular Value Decomposition (SVD) on the data matrix to obtain the left singular vector, singular value matrix, and right singular vector. Next, use the SVD results to calculate the approximate matrix of the Koopman operator and perform dimensionality reduction projection: project the approximate matrix onto the subspace spanned by the left singular vector to obtain the dimensionality-reduced matrix. Then, perform eigenvalue decomposition on the dimensionality-reduced matrix to obtain eigenvalues ​​and eigenvectors. Next, select slowly varying modes. Slowly varying modes correspond to eigenvalues ​​with moduli close to 1 (on the unit circle) and low oscillation frequencies. Then, project the initial 3D body depth image data and gait acceleration sequence data onto the linear subspace formed by the eigenvectors corresponding to these selected slowly varying eigenmodes. Finally, project the original data onto the subspace spanned by the slowly varying modes to obtain the dimensionality-reduced representation, i.e., the joint dynamics embedding set.

[0060] The pattern selection in this step is based on a physiological prior: fasting is a slow physiological process, therefore its characteristic timescale is much larger than that of fast-changing processes such as gait or respiration. Accordingly, the algorithm prioritizes feature moduli whose moduli are close to 1 or decay extremely slowly. These "slow-changing feature moduli" are interpreted as the intrinsic dynamic components that dominate slow processes such as fasting energy metabolism and center-of-mass drift. Finally, the initial 3D body surface depth image data and gait acceleration sequence data are projected onto a linear subspace formed by the feature vectors corresponding to these selected slow-changing feature moduli. The projected coordinates constitute a new, reduced-dimensional sequence, where each point linearly encodes the component of the original observation in this slow-changing dynamic subspace.

[0061] This step uses a time-scale-based pattern filtering mechanism to automatically focus on the physiological processes most relevant to fasting issues, suppressing noise from high-frequency movements or transient environmental disturbances.

[0062] Step S3: Perform filtering based on persistent coherence and knot theory on the environment-morphology-motion-centroid joint dynamic embedding set to obtain the centroid drift topological feature sequence.

[0063] Understandably, this step treats the joint dynamics embedding set as a trajectory in a high-dimensional space. The process involves dividing it into a series of continuous "trajectory segments" using a pre-defined sliding time window, each segment being considered a point cloud in a high-dimensional space. The persistent cohomology of each point cloud (i.e., the set of states of the system within a specific time window) is calculated. Persistent cohomology is achieved by constructing Vietoris-Rips complexes at different scales to track the "birth" and "death" of topological features (such as connected components and annular holes) with varying scale parameters (inter-point distance thresholds), and to record their lifetimes. Long-lived topological features are considered to reflect the system's intrinsic and stable structure. This step introduces "multi-parameter" persistent cohomology, using spatial scale parameters and environmental microclimate parameters as filtering parameters. This allows for the identification of topological features that persist at both the spatial and environmental parameter scales, effectively separating transient topological artifacts caused by environmental fluctuations. In this step, step S3 includes steps S31, S32, S33, and S34.

[0064] Step S31: Perform clustering based on the environment-morphology-motion-centroid joint dynamic embedding set to obtain a set of instantaneous attractor point clouds representing the instantaneous dynamic state of the giant panda;

[0065] Understandably, this step performs DBSCAN clustering on the "environment-morphology-motion-centroid joint dynamic embedding set," resulting in multiple clusters. Clusters with a density less than a preset threshold are deleted, and each remaining cluster is used as an instantaneous attractor point cloud representing the instantaneous dynamic state of the giant panda. This step does not rely on a fixed time window and can identify attractors based on the natural distribution of the state space.

[0066] Step S32: Perform multi-parameter persistent homology analysis based on the instantaneous attractor point cloud set to obtain a set of persistent topological invariants;

[0067] Understandably, for each attractor point cloud, the multi-parameter persistent cohomology analysis algorithm performs persistent cohomology analysis simultaneously in two dimensions: the first is the conventional spatial scale parameter (Euclidean distance between points), which controls the formation of simple complexes (such as simplexes, triangles, etc.) in the point cloud; the second is the principal component values ​​of the environmental microclimate (temperature and humidity) corresponding to that instantaneous attractor point cloud, which, as an independent parameter, controls the "visibility" of topological features under different environmental conditions. On this two-parameter plane, the multi-parameter persistent cohomology analysis algorithm tracks the "birth" and "death" of each topological feature (such as connected components, ring structures). A topological feature only exists within a specific spatial scale and a specific range of environmental parameters. Persistence is defined here as the "lifetime" of the feature on the two-parameter plane, that is, the range within which it can persist in both parameter directions.

[0068] This step clearly distinguishes the origin of topological features. For example, a ring structure that only appears under very specific environmental conditions (such as high temperature and humidity) and disappears rapidly as the environment changes indicates that it is likely induced by environmental fluctuations rather than by the system's intrinsic dynamics. Conversely, a topological feature that persists on a spatial scale and is insensitive to changes in environmental parameters (i.e., has a long "lifetime" in the environmental parameter dimension) is considered to reflect the system's inherent stability. These topological invariants (such as Betti numbers, i.e., the number of connected components, the number of loops, etc.) that exhibit high persistence in the two-parameter space are selected by using preset thresholds.

[0069] Step S33: Perform trajectory entanglement analysis based on knot theory based on the set of persistent topological invariants. By extracting the subspace trajectory spanned by the centroid coordinates separately from the joint dynamics embedding set, determine the topological invariant vector that characterizes the essential features of centroid drift.

[0070] Understandably, this step first extracts the trajectory within a subspace composed of the center of mass's state coordinates (three-dimensional spatial coordinates) from the high-dimensional "environment-morphology-motion-center of mass joint dynamic embedding set." This trajectory describes how the center of mass moves over time in its own state space. Next, the process treats this continuous trajectory as a string in high-dimensional space. Then, it calculates the knot invariants of this "string," including the self-entanglement number and the link number. The self-entanglement number quantifies the complexity of the trajectory's knotting and entanglement. A simple, smoothly moving center of mass trajectory will have a self-entanglement number close to zero, while a center of mass that moves irregularly forward, backward, left, and right due to internal mass migration will exhibit a more complex self-entanglement, resulting in a larger self-entanglement number. The link number quantifies the spatial correlation between the center of mass trajectory and a long-lived topological loop structure. For example, the center of mass trajectory may repeatedly "enter" a topological loop representing a specific energy metabolism state; the link number characterizes the tightness of this mutual entanglement. These calculated knot invariants (number of self-entanglements, number of links, etc.) then form a vector of topological invariants. This transforms the physical phenomenon of centroid drift into a stable and computable topological feature, providing crucial and robust structured information for ultimately achieving accurate fasting weight inversion.

[0071] The formulas for calculating the number of self-entanglements and the number of links are shown below:

[0072] ;

[0073] Where Wr represents the self-winding number, π is the circumference ratio, C is the centroid trajectory curve, r and r' are two points on the centroid trajectory curve, and dr and dr' are the corresponding tangent vector infinitesimals;

[0074] ;

[0075] Where Lk is the number of links, C1 is the topological loop curve identified in persistent cohomology, r1 is a point of the topological loop curve identified in persistent cohomology, and dr1 is the corresponding tangent vector element of r1.

[0076] Step S34: Based on the topological invariant vector representing the essential characteristics of centroid drift, perform anti-interference topological feature sequence synthesis processing, and arrange the synthesized centroid drift topological feature values ​​in chronological order to obtain the centroid drift topological feature sequence.

[0077] This step first designs a synthesis function that weights the norm (overall complexity) of the knot invariant vector calculated for each time window with the lifetime of the topological features within that window. For example, a complex centroid path with a high entanglement number will have its importance reduced if its corresponding topology is sensitive to environmental fluctuations (short lifetime); conversely, a centroid drift pattern that is stable under various environmental conditions, even with slightly lower absolute complexity, will have its weight increased. This step uses the stability of the topological features themselves as weights to evaluate the reliability and importance of the complexity of the centroid drift path. Finally, a scalar value, the "centroid drift topological feature value," is synthesized for each time window, and a centroid drift topological feature sequence is generated by arranging them in chronological order. The synthesis function is shown below:

[0078] ;

[0079] Where f(t) is the synthesized topological eigenvalue at time t, K(t) is the knot invariant vector at time t, and L env (t) represents the lifetime of the topological feature at time t. Let τ denote the L2 norm, τ be the loop variable that iterates through all possible time points within the time window, and T be the half width of the time window.

[0080] This step generates a time-series sequence of centroid drift topological features. This sequence is a one-dimensional time series; its values ​​comprehensively reflect the intensity and pattern complexity of centroid drift, while its trend over time clearly depicts the dynamic process of centroid drift during fasting. This sequence serves as the direct input for subsequent high-order fusion with multimodal data such as thermal and metabolic data.

[0081] Step S4: Perform environment-inertia weighted hypergraph fusion and joint state symbolization processing based on the centroid drift topological feature sequence and the multimodal data to obtain a discretized centroid-energy joint state probability measure;

[0082] The innovation of this step lies in the construction of environment-inertia weighted hyperedges. A hyperedge can connect multiple nodes appearing within the same time window (e.g., a specific centroid topological feature, a specific heat dissipation pattern, a specific metabolic level, and the current environmental state). The weight of this hyperedge is not arbitrarily set, but determined by the product of two factors: first, environmental similarity, i.e., whether the environmental signals corresponding to all nodes within the hyperedge are at similar levels; second, consistency of inertial change, i.e., whether the changing trend of the centroid topological feature is physically consistent with the changing trends of heat and metabolism (e.g., whether centroid forward drift is accompanied by increased metabolism). This weighting mechanism forces the hypergraph to learn association patterns that are within a specific environmental context and conform to physical laws, thus significantly improving the physiological credibility of the model. In this step, step S4 includes steps S41, S42, S43, and S44.

[0083] Step S41: Based on the centroid drift topological feature sequence, the body surface thermal infrared distribution time series and the metabolic label time series, perform environment-inertial weighted hypergraph construction processing. Here, by defining the centroid drift topological features, thermal infrared distribution principal components, metabolic label values ​​and environmental context driving signals at each time point as hypergraph nodes, a dynamic weighted hypergraph structure that can simultaneously characterize the coupling relationship between environmental constraints and centroid dynamics is obtained.

[0084] This step, as we understand it, instantiates the plutonium feature sequence, the thermal infrared distribution time series, and the metabolic marker time series at each time point into nodes of a hypergraph. For each time point, this step generates a hyperedge connecting the four nodes mentioned above at that moment. The weight of this hyperedge is not set to a fixed value of 1 or simply assigned a value, but is jointly determined by the environmental similarity weight and the inertial change consistency weight. The environmental similarity weight refers to calculating the cosine similarity between the environmental driving signal at the current time point and the environmental signals within its neighboring time windows. This ensures that the multivariate relationships connected by hyperedges with higher weights occur in similar environmental contexts, highlighting the constraints of environmental context on multimodal associations. The inertial change consistency weight refers to calculating the correlation or consistency (mutual information) between the rate of change of the centroid drift topological features (first-order difference) and the rate of change of the thermal infrared principal components and metabolic marker values. This forces the hypergraph to consider whether the centroid drift and thermal metabolic changes are physically and logically coordinated when learning associations. For example, does a dramatic centroid forward drift accompany a synchronous increase in metabolic rate? Ultimately, the weight of the hyperedge at that time point is the product of these two weights. By using a sliding time window, a weighted hyperedge is constructed for each moment of the entire time series, thus forming a dynamically weighted hypergraph structure.

[0085] Step S42: Perform multifractal spectrum analysis on the dynamic weighted hypergraph structure, wherein the box counting method is used to calculate the generalized fractal dimension spectrum of the weighted hypergraph at a preset scale to obtain the multifractal spectrum vector characterizing the multimodal higher-order coupling strength.

[0086] Understandably, this step employs a variant of box counting to calculate the generalized fractal dimension spectrum of the hypergraph. Classical box counting is used to calculate the fractal dimension of geometric objects; its basic idea is to cover the object with "boxes" of different sizes and observe how the required number of boxes changes as the box size decreases. For weighted hypergraphs, the "boxes" here are defined as a coarse-grained partition of the hypergraph structure. This step first introduces a variable weight threshold, then "covers" the hypergraph, i.e., finds a set of hyperedges such that the weight of each hyperedge in the set is greater than the threshold. For each given weight threshold (equivalent to a "scale"), a corresponding "quality" or "measure" (the sum of the weights of all hyperedges whose weights exceed the weight threshold) can be calculated. By systematically varying the weight threshold (i.e., changing the "scale") and observing how this "measure" changes with the weight threshold, a series of generalized fractal dimensions can be calculated. This series of generalized fractal dimensions constitutes the multifractal spectrum.

[0087] This step transforms the complex dynamic weighted hypergraph structure into a relatively low-dimensional multifractal spectral vector with clear physical meaning. This vector no longer focuses on the specific connection details of the network, but extracts the overall strength and heterogeneity characteristics of multimodal high-order coupling from a macroscopic perspective of "scale-inhomogeneity".

[0088] Step S43: Perform joint state symbolization processing based on the multifractal spectrum vector to obtain the joint state symbol sequence of the multifractal spectrum vector;

[0089] Understandably, this step first involves interpreting and discretizing the key features of the multifractal spectral vectors. Specifically, the spectral width, an indicator of the non-uniformity of coupling strength, is categorized into discrete levels such as "strong coupling" or "weak coupling" based on its numerical value. A larger spectral width indicates the simultaneous existence of extremely strong local coupling and extremely weak coupling within the system, resulting in more intense and non-uniform overall interactions.

[0090] The extreme points of the spectrum are mapped to states such as "centroid forward drift," "centroid stable," "metabolic activity," and "metabolic stability." Next, these discretized features are combined with the underlying physical states to generate a composite joint state symbol. For example, a pattern of "broad spectrum + centroid topological features indicating forward drift + high metabolic markers" is symbolized as a joint state of "strong coupling - forward drift - high metabolism." This symbol is not merely a label, but a condensation of a physical hypothesis, representing: "At this moment, the centroid has significantly shifted forward, metabolic levels are high, and there is a strong and uneven synergistic change between these mechano-energy modes." This symbolization provides operational and semantically clear state units for subsequent probabilistic inference. This allows the system to process physiological states like a language model processes words, enabling the application of mature sequence analysis techniques (such as state transition models) to study the dynamic changes in fasting body weight. It also ensures that the generated states are physiologically interpretable, rather than being a "black box" output.

[0091] Step S44: Perform short-time probability measure calculation based on weighted frequency statistics on the joint state symbol sequence to obtain a discretized centroid-energy joint state probability measure.

[0092] Understandably, this step first performs frequency statistics on the joint state symbol sequence within a short sliding time window. For example, within a window containing a positive integer number of time points, the number of occurrences of each unique joint state symbol (such as "strong coupling-drift-high metabolism") is counted. However, not all occurrences of a symbol are counted as 1; instead, each occurrence is assigned a weight proportional to the lifetime value of the topological feature corresponding to the moment the symbol appears. A higher lifetime value indicates that the topological feature generating the state symbol is more stable under environmental fluctuations, meaning the reliability of the state judgment is higher and its anti-interference ability is stronger. Therefore, the "weighted occurrence count" of a state symbol within the window is the integral of its original occurrence count and the lifetime weight corresponding to each occurrence. Subsequently, the processing normalizes the weighted occurrence counts of all state symbols within the window, i.e., by dividing by the sum of the weighted occurrence counts of all symbols. In this way, each state symbol obtains a probability value between 0 and 1 within the current time window, and the probability values ​​of all state symbols constitute a probability distribution, i.e., a short-time probability measure. This metric clearly indicates the probability that a giant panda is in various "center-of-mass-energy" joint states within the current short time window.

[0093] This step transforms the discrete symbol sequence into a continuous, time-evolving centroid-energy joint state probability measure time series. This probability measure forms the basis for subsequent Bayesian inversion; it encompasses both the statistical regularities of state occurrence and incorporates a measure of uncertainty in state estimation, providing crucial and information-rich probabilistic input for ultimately achieving highly robust dynamic calibration of fasting body weight.

[0094] Step S5: Perform quantum dual-interference compensation inversion processing based on the discretized centroid-energy joint state probability measure to obtain the dynamic calibration result of the fasting weight of the giant panda after compensating for centroid drift and environmental fluctuation effects.

[0095] Understandably, this step involves precise disturbance compensation and final inversion using the centroid-energy joint state probability measure, resulting in a precisely calibrated estimate of fasting body weight. Its processing method is highly innovative, elevating the classical physiological state probability distribution to a quantum state description and utilizing unique mechanisms from quantum theory to simulate and correct two key disturbances. In this step, step S5 includes steps S51, S52, S53, and S54.

[0096] Step S51: Based on the discretized centroid-energy joint state probability measure, quantum state preparation of energy level encoding is performed to obtain the quantum state representation of the fasting physiological system of the giant panda.

[0097] Understandably, this step first prepares the discrete joint state probability measures into a quantum superposition state. By mapping each discrete joint state (such as "stable-low metabolism") to an abstract, orthogonal quantum ground state (i.e., defining each ground state as corresponding to a discrete center-of-mass-energy joint state), a complete basis of the state space is constructed. The square root of each probability value is assigned to the corresponding ground state as its probability amplitude. Based on this, the state of the entire physiological system is no longer a classical probability distribution, but is represented as a linear superposition of all these ground states. This superposition state is the quantum state representation of the giant panda's fasting physiological system.

[0098] The quantum state representation of the fasting physiological system of giant pandas is the object and foundation of all subsequent quantization processes (such as phase rotation compensation and error correction). It places the traditional statistical inference results within a framework that can utilize unique effects such as quantum interference and entanglement, thus laying the foundation for achieving high-precision "interference compensation" inversion.

[0099] Step S52: Based on the quantum state representation of the fasting physiological system of the giant panda, perform phase compensation processing for the centroid drift to obtain the processed quantum state representation;

[0100] It is understandable that the center-of-mass drift (e.g., forward drift, backward drift) in this step leads to a systematic error related to the center-of-mass position between the morphology-based estimated weight and the actual fasting weight. In the quantum model, this systematic bias bound to a specific state is cleverly modeled as applying a specific phase rotation to the probability amplitude corresponding to that state. For example, if the state corresponds to "center-of-mass forward drift," then the probability amplitude of that state is multiplied by a phase factor, where the phase angle is proportional to the topological eigenvalue of the center-of-mass drift. This is equivalent to rotating the probability amplitude vector of that state by an angle in the complex plane.

[0101] Mathematically, this operation is described by a phase rotation operator, which evolves the quantum state into a new state containing a phase deviation. The subsequent phase compensation process involves applying an exact conjugate inverse phase rotation, that is, multiplying the probability amplitude of the state by a phase factor. According to the principles of quantum mechanics, the product of two conjugate phase factors equals 1, thus canceling out the phase shift introduced by the center-of-mass drift model.

[0102] Step S53: Perform anti-interference filtering processing on the processed quantum state representation to obtain the filtered quantum state representation;

[0103] Understandably, this step first performs time-frequency analysis on the time series of quantum state probability amplitudes, decomposing the signal into the frequency domain using short-time Fourier transform or wavelet transform to identify low-frequency components related to physiological processes (reflecting actual fasting state changes) and high-frequency noise components related to environmental fluctuations. Based on this analysis, this application designs an adaptive Wiener filter whose filtering parameters are dynamically adjusted by the intensity of environmental fluctuations. Specifically, fluctuation intensity indicators (such as the variance of temperature and humidity) are calculated in real-time from environmental microclimate field data. When environmental fluctuations are severe, the filtering intensity is increased to effectively suppress noise; when the environment is relatively stable, the filtering intensity is decreased to retain more physiological details. This filter weights the probability amplitude spectrum of the quantum state in the frequency domain, retaining low-frequency physiological signals and suppressing high-frequency environmental noise, resulting in a filtered quantum state representation.

[0104] The adaptive Wiener filter is shown below:

[0105] ;

[0106] Where H(f,t) is the transfer function of the filter at frequency f, and P s (f,t) represents the signal power spectral density, P n (f,t) is the noise power spectral density, and λ(E(t)) is the environmental adaptive adjustment function, which is shown below:

[0107] ;

[0108] Where λ(E(t)) is the environmental adaptive adjustment function, λ min λ represents the minimum filtering strength. max The maximum value of the filter intensity is σ, where σ is the sigmoid function and E(t) is the environmental fluctuation intensity index (temperature and humidity variance). th ΔE represents the environmental fluctuation threshold, and ΔE represents the adjustment sensitivity parameter.

[0109] Step S54: Based on the filtered quantum state representation, perform weight eigenvalue inversion processing based on quantum Bayes inference to obtain the dynamic calibration result of the fasting weight of the giant panda.

[0110] Understandably, this step first assigns a preset weight eigenvalue to the ground state of each "center-of-mass-energy" joint state. These eigenvalues ​​are learned through historical calibration data and reflect the correspondence between different physiological states and weight. Subsequently, quantum measurements are performed on the filtered quantum states. According to the principles of quantum mechanics, the measurement results will probabilistically collapse to a certain ground state. The final fasting weight estimate is the statistical expectation of all possible measurement results, as shown below:

[0111] ;

[0112] Where W represents the expected value, and n is the number of all possible states. Let w be the posterior probability after all the aforementioned compensation and filtering processes. j It is the prior eigenvalue of body weight.

[0113] Meanwhile, this application can calculate the variance of the estimated value as a measure of uncertainty, thereby obtaining the weight estimate and its variance, and using it as a dynamic calibration result.

[0114] Example 2:

[0115] like Figure 2 As shown, this embodiment provides a multimodal data fusion-based dynamic calibration system for fasting weight of giant pandas. (See also...) Figure 2 The system includes an acquisition unit 701, a processing unit 702, a filtering unit 703, a fusion unit 704, and an inversion unit 705.

[0116] The acquisition unit 701 is used to acquire multimodal data of giant pandas in their natural state. The multimodal data includes three-dimensional body surface depth image data at multiple times, gait acceleration sequence, body surface thermal infrared distribution time series, event timestamp sequence of feeding behavior, metabolic marker time series before and after each feeding, trunk inertial measurement data, and spatiotemporal sequence of local environmental climate field.

[0117] Processing unit 702 is used to perform environment context-aware Koopman-centroid dynamics joint embedding processing based on the multimodal basic data to obtain an environment-morphology-motion-centroid joint dynamics embedding set.

[0118] The filtering unit 703 is used to perform filtering based on persistent cohomology and knot theory according to the environment-morphology-motion-centroid joint dynamic embedding set to obtain the centroid drift topological feature sequence.

[0119] The fusion unit 704 is used to perform environment-inertia weighted hypergraph fusion and joint state symbolization processing based on the centroid drift topological feature sequence and the multimodal data to obtain a discretized centroid-energy joint state probability measure.

[0120] The inversion unit 705 is used to perform quantum dual-interference compensation inversion processing based on the discretized centroid-energy joint state probability measure to obtain the dynamic calibration result of the fasting weight of the giant panda after compensating for centroid drift and environmental fluctuation effects.

[0121] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0122] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0123] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for dynamic calibration of fasting weight of giant pandas using multimodal data fusion, characterized in that, include: Acquire multimodal data of giant pandas in their natural state. The multimodal data includes three-dimensional body surface depth image data at multiple time points, gait acceleration sequence, body surface thermal infrared distribution time series, event timestamp sequence of feeding behavior, metabolic marker time series before and after each feeding, trunk inertial measurement data, and spatiotemporal sequence of local environmental climate field. Based on the multimodal data, a context-aware Koopman-centroid dynamics joint embedding process is performed to obtain an environment-morphology-motion-centroid joint dynamics embedding set; Based on the environment-morphology-motion-centroid joint dynamic embedding set, filtering based on persistent cohomology and knot theory is performed to obtain the centroid drift topological feature sequence. Based on the centroid drift topological feature sequence and the multimodal data, environmental-inertial weighted hypergraph fusion and joint state symbolization are performed to obtain a discretized centroid-energy joint state probability measure. Based on the discretized centroid-energy joint state probability measure, a quantum dual-interference compensation-based inversion process was performed to obtain the dynamic calibration results of the fasting weight of giant pandas, which compensated for centroid drift and environmental fluctuation effects.

2. The method for dynamic calibration of fasting weight of giant pandas based on multimodal data fusion according to claim 1, characterized in that, Based on the multimodal data, a context-aware Koopman-centroid dynamics joint embedding process is performed, including: Environmental driving factors are extracted based on the spatiotemporal sequence of the local environmental climate field. Principal component analysis is used to reduce the dimensionality and extract the principal component sequence that changes over time. Then, mutual information analysis is performed between the principal component sequence and the behavioral event timestamp sequence. Environmental principal components with mutual information greater than a preset threshold with eating and activity behaviors are selected as environmental context driving signals. Based on the three-dimensional body surface depth image data and inertial measurement unit data, a centroid dynamic system is constructed. By treating the three-dimensional body surface point cloud of continuous frames as a time-varying massless shell, the net external force and net torque acting on the shell are estimated by inverting the Newton-Euler equation using the torso inertial measurement unit data. The shell is then discretized into multiple rigid segments and assigned equivalent mass based on anatomical priors to obtain a centroid dynamic model that reflects the internal mass transfer. Based on the environmental context driving signal and the centroid dynamics model, the extended phase space of the environment-centroid coupling is constructed. The environmental driving signal is introduced as an external forcing term into the state equation of the centroid dynamics model, and the extended phase space of the environment-centroid coupling relationship is constructed based on the state equation and the gait acceleration sequence. Based on the extended phase space, the dynamic mode decomposition algorithm is applied to solve the approximate matrix of its Koopman operator. The matrix is ​​then subjected to spectral decomposition. The eigenvectors corresponding to the slow-varying eigenmodes most relevant to the fasting process are selected. The three-dimensional body surface depth image data and gait acceleration sequences at multiple time points are projected onto the subspace spanned by these eigenvectors to obtain the final environment-morphology-motion-centroid joint dynamic embedding set.

3. The method for dynamic calibration of fasting weight of giant pandas based on multimodal data fusion according to claim 1, characterized in that... The filtering process based on persistent cohomology and knot theory is performed according to the aforementioned environment-morphology-motion-centroid joint dynamic embedding set, including: Clustering is performed based on the aforementioned environment-morphology-motion-centroid joint dynamic embedding set to obtain a set of instantaneous attractor point clouds representing the instantaneous dynamic state of the giant panda; Based on the instantaneous attractor point cloud set, multi-parameter persistent homology analysis is performed to obtain a set of persistent topological invariants; Based on the set of persistent topological invariants, trajectory entanglement analysis based on knot theory is performed. By extracting the subspace trajectory spanned by the centroid coordinates separately from the joint dynamics embedding set, the topological invariant vector characterizing the essential features of centroid drift is determined. Based on the topological invariant vector that characterizes the essential features of centroid drift, an anti-interference topological feature sequence synthesis process is performed, and the synthesized centroid drift topological feature values ​​are arranged in chronological order to obtain the centroid drift topological feature sequence.

4. The method for dynamic calibration of fasting weight of giant pandas based on multimodal data fusion according to claim 1, characterized in that... Based on the centroid drift topological feature sequence and the multimodal data, environment-inertia weighted hypergraph fusion and joint state symbolization processing are performed, including: Based on the centroid drift topological feature sequence, the body surface thermal infrared distribution time series and the metabolic label time series, an environment-inertial weighted hypergraph construction process is performed. In this process, the centroid drift topological features, thermal infrared distribution principal components, metabolic label values ​​and environmental context driving signals at each time point are defined together as hypergraph nodes, resulting in a dynamic weighted hypergraph structure that can simultaneously characterize the coupling relationship between environmental constraints and centroid dynamics. Based on the multifractal spectrum analysis on the dynamic weighted hypergraph structure, the generalized fractal dimension spectrum of the weighted hypergraph at a preset scale is calculated using the box counting method to obtain the multifractal spectrum vector characterizing the multimodal higher-order coupling strength. Based on the multifractal spectral vector, a joint state symbolization process is performed to obtain the joint state symbol sequence of the multifractal spectral vector; Based on the joint state symbol sequence, a short-time probability measure calculation based on weighted frequency statistics is performed to obtain a discretized centroid-energy joint state probability measure.

5. The method for dynamic calibration of fasting weight of giant pandas based on multimodal data fusion according to claim 1, characterized in that... Based on the discretized centroid-energy joint state probability measure, a quantum dual-interference compensation-based inversion process is performed, including: Quantum state preparation based on energy level encoding using discretized centroid-energy joint state probability measure yields quantum state representation of the fasting physiological system of giant panda. Based on the quantum state representation of the fasting physiological system of the giant panda, phase compensation processing is performed to address the centroid drift, resulting in the processed quantum state representation. Based on the processed quantum state representation, an anti-interference filtering process is performed to target environmental fluctuations, resulting in a filtered quantum state representation. Based on the filtered quantum state representation, the weight eigenvalue inversion process based on quantum Bayes inference is performed to obtain the dynamic calibration result of the fasting weight of the giant panda.

6. A dynamic calibration system for fasting weight of giant pandas using multimodal data fusion, characterized in that, include: The acquisition unit is used to acquire multimodal data of giant pandas in their natural state. The multimodal data includes three-dimensional body surface depth image data at multiple times, gait acceleration sequence, body surface thermal infrared distribution time series, event timestamp sequence of feeding behavior, metabolic marker time series before and after each feeding, trunk inertial measurement data, and spatiotemporal sequence of local environmental climate field. The processing unit is used to perform environment context-aware Koopman-centroid dynamics joint embedding processing based on the multimodal data to obtain an environment-morphology-motion-centroid joint dynamics embedding set. The filtering unit is used to perform filtering based on persistent cohomology and knot theory according to the environment-morphology-motion-centroid joint dynamic embedding set to obtain the centroid drift topological feature sequence. The fusion unit is used to perform environment-inertia weighted hypergraph fusion and joint state symbolization processing based on the centroid drift topological feature sequence and the multimodal data to obtain a discretized centroid-energy joint state probability measure. The inversion unit is used to perform quantum dual-interference compensation inversion processing based on the discretized centroid-energy joint state probability measure to obtain the dynamic calibration result of the fasting weight of giant pandas that has been compensated for centroid drift and environmental fluctuation effects.

7. The multimodal data fusion dynamic calibration system for fasting weight of giant pandas according to claim 6, characterized in that, The processing unit includes: The first processing subunit is used to extract environmental driving factors based on the spatiotemporal sequence of the local environmental climate field. Specifically, it performs dimensionality reduction and extracts the principal component sequence that changes over time through principal component analysis, and then performs mutual information analysis with the behavioral event timestamp sequence to screen out environmental principal components with mutual information greater than a preset threshold with eating and activity behaviors as environmental context driving signals. The second processing subunit is used to construct a centroid dynamic system based on the three-dimensional body surface depth image data and inertial measurement unit data. By treating the three-dimensional body surface point cloud of continuous frames as a time-varying massless shell, the net external force and net torque acting on the shell are estimated by inverting the Newton-Euler equation using the torso inertial measurement unit data. The shell is then discretized into multiple rigid segments and given equivalent mass based on anatomical priors to obtain a centroid dynamic model that reflects the internal mass transfer. The third processing subunit is used to construct an extended phase space of the environment-center-of-mass coupling based on the environmental context driving signal and the center-of-mass dynamics model. The extended phase space of the environment-center-of-mass coupling relationship is obtained by introducing the environmental driving signal as an external forcing term into the state equation of the center-of-mass dynamics model and constructing the state equation and gait acceleration sequence. The fourth processing subunit is used to apply the dynamic mode decomposition algorithm based on the extended phase space to solve the approximate matrix of its Koopman operator, and to perform spectral decomposition on the matrix. It selects the eigenvectors corresponding to the slow-varying eigenmodes most relevant to the fasting process, and projects the three-dimensional body surface depth image data and gait acceleration sequences at multiple times onto the subspace spanned by these eigenvectors to obtain the final environment-morphology-motion-centroid joint dynamic embedding set.

8. The multimodal data fusion dynamic calibration system for fasting weight of giant pandas according to claim 6, characterized in that, The filtering unit includes: The first filtering subunit is used to perform clustering based on the environment-morphology-motion-centroid joint dynamic embedding set to obtain a set of instantaneous attractor point clouds representing the instantaneous dynamic state of the giant panda; The second filtering subunit is used to perform multi-parameter persistent homology analysis based on the instantaneous attractor point cloud set to obtain a set of persistent topological invariants. The third filtering subunit is used to perform trajectory entanglement analysis based on knot theory based on the set of persistent topological invariants. It determines the topological invariant vector that characterizes the essential features of centroid drift by extracting the subspace trajectory spanned by the centroid coordinates separately from the joint dynamics embedding set. The fourth filtering subunit is used to perform anti-interference topological feature sequence synthesis processing based on the topological invariant vector that characterizes the essential features of centroid drift, and to arrange the synthesized centroid drift topological feature values ​​in chronological order to obtain the centroid drift topological feature sequence.

9. The multimodal data fusion dynamic calibration system for fasting weight of giant pandas according to claim 6, characterized in that, The fusion unit includes: The first fusion subunit is used to perform environment-inertial weighted hypergraph construction processing based on the centroid drift topological feature sequence, the body surface thermal infrared distribution time series and the metabolic label time series. Here, by defining the centroid drift topological features, thermal infrared distribution principal components, metabolic label values ​​and environmental context driving signals at each time point as hypergraph nodes, a dynamic weighted hypergraph structure that can simultaneously characterize the coupling relationship between environmental constraints and centroid dynamics is obtained. The second fusion subunit is used for multifractal spectrum analysis processing on the dynamic weighted hypergraph structure, wherein the box counting method is used to calculate the generalized fractal dimension spectrum of the weighted hypergraph at a preset scale to obtain the multifractal spectrum vector characterizing the multimodal higher-order coupling strength. The third fusion subunit is used to perform joint state symbolization processing based on the multifractal spectrum vector to obtain the joint state symbol sequence of the multifractal spectrum vector; The fourth fusion subunit is used to perform short-time probability measure calculation based on weighted frequency statistics based on the joint state symbol sequence to obtain a discretized centroid-energy joint state probability measure.

10. The multimodal data fusion dynamic calibration system for fasting weight of giant pandas according to claim 6, characterized in that, The inversion unit includes: The first inversion subunit is used for quantum state preparation based on the discretized centroid-energy joint state probability measure to encode the energy level, thereby obtaining the quantum state representation of the fasting physiological system of the giant panda. The second inversion subunit is used to perform phase compensation processing for centroid drift based on the quantum state representation of the fasting physiological system of the giant panda, so as to obtain the processed quantum state representation. The third inversion subunit is used to perform anti-interference filtering processing for environmental fluctuations based on the processed quantum state representation to obtain the filtered quantum state representation; The fourth inversion subunit is used to perform weight eigenvalue inversion processing based on quantum Bayes inference based on the filtered quantum state representation, and to obtain the dynamic calibration result of the fasting weight of the giant panda.

Images