Mouse weight measuring method based on artificial intelligence

Abstract

The invention discloses a mouse weight measuring method based on artificial intelligence, which comprises the following steps: acquiring a mouse multi-view dynamic image sequence through a mouse 3D behavior phenotype analysis platform and extracting related feature information, correcting phase shift deviation by using a dynamic phase shift calibration prediction model, generating three-dimensional volume associated data by using a gradient optimization volume estimation network, and determining the weight of a mouse according to the three-dimensional volume associated data. And carrying out space-time dimension calibration by using a space-time phase alignment optimization algorithm, constructing a correlation mapping model of mouse weight and multi-dimensional characteristics by combining output data of each link, and calculating and outputting a weight measurement result based on the model. Wherein the application of the model and the algorithm realizes phase shift deviation correction, accurate volume estimation and time-space data alignment, the processing effect of related steps is enhanced through hierarchical step-by-step operation, the method does not need manual intervention, the accuracy, efficiency and objectivity of weight measurement are improved, and the method is suitable for large-scale popularization and application. The method is suitable for mouse weight monitoring in the fields of biomedical research, drug research and development and the like.

Description

Technical Field

[0001] This invention relates to the field of mouse weight measurement technology, and more particularly to a method for measuring mouse weight based on artificial intelligence. Background Technology

[0002] In biomedical research, drug development, and animal behavior analysis, mice are commonly used model organisms, and their weight data is a core indicator reflecting physiological state, metabolic level, and the effectiveness of experimental interventions. Traditional mouse weight measurement relies on manual operation, requiring mice to be removed from their enclosure and placed on a weighing device to acquire data. This process is not only inefficient and unable to meet the high-frequency weight monitoring needs of large-scale experiments, but it may also cause stress responses in mice due to human intervention, affecting their normal physiological state and thus interfering with the objectivity of experimental data. With the rapid development of artificial intelligence technology in image analysis, 3D reconstruction, and other fields, weight measurement schemes based on contactless image acquisition and intelligent algorithms have gradually become a research hotspot. By integrating multi-view image acquisition, dynamic feature extraction, and intelligent model analysis, accurate and efficient measurement of mouse weight can be achieved, providing more reliable data support for related research while avoiding many limitations of manual operation.

[0003] Existing technologies have significant shortcomings in non-contact measurement of mouse weight: On the one hand, traditional image analysis methods struggle to effectively address phase shift biases and spatiotemporal data misalignments arising from dynamic mouse behavior. Simple image feature extraction and fitting calculations fail to accurately capture the impact of mouse posture changes and movement trajectory shifts on volume estimation and weight mapping, resulting in insufficient accuracy of measurement results. On the other hand, existing technologies lack dedicated intelligent model designs tailored to mouse physiological characteristics and behavioral patterns. They do not construct a full-link optimization mechanism for phase shift calibration, volume estimation, spatiotemporal alignment, and weight correlation. They often employ general-purpose algorithms for data processing, which are difficult to adapt to the small size and agile movement of mice. This results in insufficient depth of model mining of multidimensional mouse features, making it impossible to establish a stable weight correlation mapping relationship, thus affecting the applicability and reliability of the measurement method. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides an artificial intelligence-based method for determining mouse weight.

[0005] The technical solution adopted in this invention is an artificial intelligence-based method for measuring mouse weight, comprising the following steps: S1, acquiring multi-view dynamic image sequences of mice through a mouse 3D behavioral phenotypic analysis platform, and extracting feature information representing mouse posture, contour, and movement trajectory from the images; S2, calling a mouse dynamic phase shift calibration prediction model to correct phase shift deviations in the acquired feature information, and adjusting the phase shift-related deviation data in the feature information through the dynamic mapping relationship constructed by the model; S3, processing the corrected feature information using a gradient-optimized volume estimation network, and generating mouse three-dimensional volume correlation data through gradient transfer and parameter iterative updates between network layers; S4, using a spatiotemporal phase alignment optimization algorithm to perform spatiotemporal dimension calibration on the three-dimensional volume correlation data, and establishing alignment mappings of volume data under different time nodes and spatial perspectives; S5, combining the correction parameters output by the mouse dynamic phase shift calibration prediction model, the volume correlation data generated by the gradient-optimized volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm, constructing a correlation mapping model between mouse weight and multidimensional features; S6, calculating mouse weight parameters based on the correlation mapping model, and outputting the mouse weight measurement results.

[0006] Furthermore, the expression for the mouse dynamic phase shift calibration prediction model is: ,in, These are the calibrated phase shift characteristic parameters. These are the original phase shift characteristic parameters. This represents the characteristic value of mouse movement rate. The time interval between adjacent frames. These are the model weight coefficients. This is the phase shift deviation correction factor. The time decay factor, This represents the gradient value of the original phase shift characteristic parameters.

[0007] Furthermore, the expression for the gradient-optimized volume estimation network is: ,in, For the estimated mouse three-dimensional volume correlation data, For the corrected feature information components, The number of feature components, , These are the network weight parameters. For network bias terms, For activation function, For gradient optimization coefficients, This is the gradient operator.

[0008] Furthermore, the expression for the spatiotemporal phase alignment optimization algorithm is: ,in, This is the volume data after spatiotemporal alignment. For the first Weight of each time node For the first Each spatial perspective weight, For the first Phase angle at each time node, For the first A spatial perspective phase angle, The total number of spatiotemporal nodes. For the estimated three-dimensional volume correlation data of mice.

[0009] Furthermore, the expression for constructing the association mapping model between mouse body weight and multidimensional features is as follows: ,in, The results are the mouse body weight measurement results. The correlation coefficient, This is the volume data after spatiotemporal alignment. These are the calibrated phase shift characteristic parameters. The gradient values ​​for the volume estimation data. For the first The weight coefficients of each feature component, For the corrected feature information components, This represents the total number of characteristic components.

[0010] Furthermore, the association model for the feature information collected by the mouse 3D behavioral phenotype analysis platform is as follows: ,in, To collect comprehensive feature information, For the collection coefficient, For the first Image acquisition weights from each perspective, For the first Image feature matrix from each perspective For the first Feature compensation values ​​for each viewpoint This represents the total number of viewpoints collected.

[0011] Further, step S3 includes the following sub-steps: S31, the corrected feature information is hierarchically divided into multiple basic processing units according to feature dimension, data type, and correlation with volume; S32, different basic processing units are input into the corresponding layers of the gradient optimization volume estimation network, and the feature data is dimension-matched and format-converted through the network input layer to make the data meet the network computation requirements; S33, the gradient transfer mechanism inside the network is activated, the data is nonlinearly transformed through the hidden layer, and the feature information is deeply extracted by combining the parameter interaction of adjacent layers; S34, the processed feature information is integrated based on the parameter configuration of the output layer to generate three-dimensional volume correlation data that can directly reflect the volume characteristics of the mouse.

[0012] Further, step S4 includes the following sub-steps: S41, extracting the time dimension identifier and spatial perspective identifier from the three-dimensional volume association data, establishing a spatiotemporal information index table, and clarifying the spatiotemporal position corresponding to different data; S42, based on the core logic of the spatiotemporal phase alignment optimization algorithm, calculating the phase difference and deviation value of the volume data under different spatiotemporal nodes, and determining the calibration data nodes that need to be adjusted; S43, performing phase compensation and position calibration on the calibration data nodes according to the alignment rules preset by the algorithm, so that the data of different spatiotemporal dimensions achieve consistent association; S44, integrating and verifying all the calibrated volume data, removing abnormal data that does not meet the spatiotemporal alignment standard, and retaining valid data for subsequent processing.

[0013] Further, S5 includes the following sub-steps: S51, collecting the calibration parameters of the mouse dynamic phase shift calibration prediction model, the volume correlation data of the gradient-optimized volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm to establish a multidimensional dataset; S52, performing feature screening on the multidimensional dataset, retaining core influencing features and removing redundant data based on the correlation strength between the data and mouse weight; S53, constructing the basic framework of the mapping model, setting the structure of the input layer, hidden layer, and output layer of the model, and clarifying the parameter transmission paths of different levels; S54, inputting the screened core feature data into the model framework, adjusting the model parameters through data training, and determining the quantitative correlation between mouse weight and multidimensional features.

[0014] An AI-based method for determining mouse weight is implemented through several units, including: an image sequence acquisition and feature extraction unit, which acquires dynamic image sequences of the mouse from different perspectives using a pre-set image acquisition device, separates and extracts feature information representing the mouse's posture, contour, and movement trajectory from the images, and transmits the extracted feature information to a dynamic phase shift deviation correction unit; a dynamic phase shift deviation correction unit, which receives the feature information transmitted by the feature extraction unit, calls a mouse dynamic phase shift calibration prediction model to correct the phase shift deviation in the feature information, and sends the corrected feature information to a 3D volume correlation data generation unit; and a 3D volume correlation data generation unit, which receives the corrected feature information, processes the data through a gradient-optimized volume estimation network, and generates a 3D volume correlation data of the mouse. Volume correlation data is transmitted to the spatiotemporal dimension calibration unit. The spatiotemporal dimension calibration unit receives the three-dimensional volume correlation data, performs spatiotemporal dimension calibration using a spatiotemporal phase alignment optimization algorithm, establishes alignment mappings for volume data at different spatiotemporal nodes, and transmits the calibrated data to the weight correlation mapping model construction unit. The weight correlation mapping model construction unit collects the correction parameters, volume correlation data, and calibration results output by different units, constructs a correlation mapping model between mouse weight and multidimensional features, and transmits the model to the weight parameter calculation unit. The weight parameter calculation unit receives the correlation mapping model, calculates the mouse weight parameters based on the quantitative correlation relationship set in the model, and outputs the mouse weight measurement results. Different units engage in bidirectional data interaction and collaborative work through the data transmission interface.

[0015] Beneficial Effects: This invention proposes an artificial intelligence-based method for measuring mouse weight. It utilizes a 3D behavioral phenotypic analysis platform to acquire dynamic features from multiple perspectives. Combined with a dynamic phase shift calibration prediction model, a gradient-optimized volume estimation network, and a spatiotemporal phase alignment optimization algorithm, it forms a complete technical solution encompassing feature extraction, bias correction, volume estimation, and spatiotemporal calibration. This effectively solves the problems of dynamic phase shift bias and spatiotemporal data misalignment that are difficult to handle with existing technologies. Through deep adaptation of mouse behavioral characteristics using a dedicated model, it accurately captures the correlation between posture changes and volume and weight, significantly improving measurement accuracy. Simultaneously, the entire measurement process requires no manual intervention, avoiding the stress response and data interference caused by traditional manual weighing. Furthermore, the automated data processing capabilities of the intelligent algorithm meet the high-frequency weight monitoring needs of large-scale experiments, overcoming the limitations of low efficiency and insufficient applicability of existing technologies. This provides a more reliable and efficient means of weight measurement for biomedical research, drug development, and other fields, balancing data objectivity and experimental convenience, and promoting the standardization and large-scale development of mouse-related research. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method steps of the present invention;

[0017] Figure 2 This is a unit diagram for implementing the method of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] like Figure 1 As shown, the artificial intelligence-based method for determining mouse weight includes the following steps:

[0020] S1. A multi-view dynamic image sequence of mice was collected through a mouse 3D behavioral phenotyping platform, and feature information representing mouse posture, contour and movement trajectory was extracted from the images.

[0021] Specifically, step S1 involves acquiring a multi-view dynamic image sequence of mice using a mouse 3D behavioral phenotypic analysis platform. The platform is equipped with eight evenly distributed high-definition industrial cameras, with a resolution of 3840×2160 pixels, a frame rate of 60 frames per second, and a continuous acquisition duration of 30 seconds, covering various behavioral states of the mice, including stillness, movement, and turning. During acquisition, the cameras are synchronously triggered to ensure that the synchronization error of image frames from each viewpoint is controlled within 10 milliseconds. The platform's built-in image acquisition control module receives image data transmitted from each camera in real time and stores it on a local server. Subsequently, feature information representing the mouse's posture, contour, and movement trajectory is extracted from the images based on image segmentation technology. Posture features include 12 key indicators such as trunk bending angle, limb extension degree, and head orientation. Contour features are extracted using an edge detection algorithm to obtain the coordinates of 100 feature points on the mouse's outer contour. Movement trajectory features are calculated based on feature point matching between adjacent frames to determine the mouse's center of gravity movement path and speed change. All extracted feature information is integrated according to a unified data format to form a feature matrix with dimensions of 128×30×60, providing complete raw data support for subsequent processing.

[0022] S2, call the mouse dynamic phase shift calibration prediction model to correct the phase shift deviation of the collected feature information, and adjust the deviation data related to phase shift in the feature information through the dynamic mapping relationship constructed by the model;

[0023] Specifically, step S2 calls the mouse dynamic phase shift calibration prediction model to correct phase shift bias in the collected feature information. This model is based on a deep neural network and includes an input layer, four hidden layers, and an output layer. The number of nodes in the input layer is set to 128, the number of nodes in the hidden layers are 256, 512, 512, and 256 respectively, and the number of nodes in the output layer is 64. During model training, 3000 sets of mouse multi-view image feature data including phase shift bias annotations are used as the training set. The number of training iterations is set to 1000, the initial learning rate is 0.001, and it decreases by 50% every 200 iterations. During calibration, the feature matrix extracted in step S1 is input into the model and passed through the input layer to the hidden layer. Each hidden layer performs a nonlinear transformation on the feature data through an activation function. Combined with the dynamic mapping relationship learned during model training, the bias data related to phase shift in the feature information is adjusted. The focus is on correcting the phase shift bias caused by differences in camera shooting angle and changes in mouse movement speed. During the calibration process, the bias adjustment range of the model output is controlled between -0.15 and 0.15. The phase shift correction of the entire data is completed by traversing the feature matrix frame by frame, generating calibrated feature information with the same dimension as the original feature matrix to ensure the accuracy of subsequent data processing.

[0024] S3 uses a gradient-optimized volume estimation network to process the corrected feature information. Through gradient transfer and parameter iterative update between network layers, mouse three-dimensional volume correlation data is generated.

[0025] Specifically, step S3 uses a gradient-optimized volume estimation network to process the corrected feature information. This network includes a feature enhancement module, a gradient calculation module, and a volume-related data generation module. The feature enhancement module has six convolutional layers with kernel sizes of 3×3 and 5×5 alternating, and a stride of 1 for each layer. The convolution operation enhances the expressive power of the feature information. The gradient calculation module is based on the backpropagation algorithm to calculate the gradient values ​​of the feature data at each level. The gradient threshold is set to 0.005. When the gradient value is lower than the threshold, the parameters are updated. The volume-related data generation module includes two fully connected layers with 128 and 64 nodes, respectively. During processing, the corrected feature information is first input into the feature enhancement module. After convolution and pooling, deeper volume-related features are extracted. Subsequently, the gradient calculation module solves the gradient of the enhanced feature data. Through gradient transfer between network layers, the correlation gradient between feature data and volume estimation results is captured. Combined with the preset parameter iteration update rules, the network parameters are updated every 50 frames of data. During the iteration, the parameter update step size is dynamically adjusted according to the gradient value, ranging from 0.001 to 0.003. Finally, the volume correlation data generation module integrates the processed feature data to generate multi-dimensional three-dimensional volume correlation data, including mouse trunk volume, limb volume, head volume, etc. The data dimension is set to 64×30. Each data point corresponds to the volume correlation value of a certain part of the mouse at a specific time frame, providing basic data for subsequent spatiotemporal calibration.

[0026] S4. The spatiotemporal phase alignment optimization algorithm is used to perform spatiotemporal dimension calibration on the three-dimensional volume correlation data, and establish the alignment mapping of volume data under different time nodes and spatial perspectives.

[0027] Specifically, step S4 uses a spatiotemporal phase alignment optimization algorithm to perform spatiotemporal dimension calibration on the three-dimensional volume correlation data. The algorithm first sets the time calibration window size to 10 frames, and the spatial view calibration weight is allocated based on the shooting angle of each camera. The front and side view weights are set to 0.2, and the oblique view weights are set to 0.1 to ensure the balanced participation of data from each view. During the calibration process, the three-dimensional volume correlation data is first processed in the temporal dimension. A sliding time calibration window with a step size of 5 frames is used to calculate the phase value of each frame within the window. The time calibration coefficient is calculated based on the phase difference, and phase compensation is performed on the volume data at different time points to keep the phase difference between adjacent frames within 5°. Then, spatial dimension calibration is performed. The corresponding volume data at each viewpoint is extracted, the phase offset of the data at different viewpoints is calculated, and the offset data is adjusted by a linear interpolation algorithm to establish the phase alignment relationship between spatial viewpoints. Finally, a spatiotemporal alignment mapping matrix is ​​constructed with a matrix dimension of 64×30×8. The time-calibrated volume data and the spatially calibrated viewpoint data are fused to clarify the correspondence between volume data at different time points and spatial viewpoints. This ensures that the calibrated volume data maintains consistency in the spatiotemporal dimension, eliminates data misalignment caused by mouse movement and camera viewpoint differences, and provides unified standard data support for the subsequent construction of the correlation mapping model.

[0028] S5. Combining the correction parameters output by the mouse dynamic phase shift calibration prediction model, the volume correlation data generated by the gradient optimization volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm, a correlation mapping model between mouse body weight and multidimensional features is constructed.

[0029] Specifically, step S5 combines the correction parameters output by the mouse dynamic phase shift calibration prediction model, the volume correlation data generated by the gradient-optimized volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm to construct a correlation mapping model between mouse weight and multidimensional features. First, the 64-dimensional correction parameters output by the model, the 64×30-dimensional volume correlation data generated by the network, and the 64×30×8-dimensional spatiotemporal calibration data output by the algorithm are collected. The three types of data undergo dimensional unification processing, and a comprehensive feature matrix with dimensions of 64×30×10 is formed by data concatenation. Then, a feature selection algorithm is used to screen core features strongly correlated with mouse weight, setting a correlation threshold of 0.7, retaining 48 core feature dimensions that meet the conditions, and removing redundant data. The model construction adopts a multilayer perceptron structure, with an input layer, three hidden layers, and an output layer. The number of nodes in the input layer is 48×30×10, the number of nodes in the hidden layers are 1024, 512, and 256 respectively, and the number of nodes in the output layer is 1. During model training, 2000 sets of measured mouse weight data and corresponding comprehensive feature data were used as training samples. The training batch size was set to 32, and the number of iterations was 800. The model prediction error was calculated using the mean squared error loss function, and the model parameters were adjusted by combining the adaptive gradient descent algorithm to gradually converge the deviation between the model prediction value and the measured value. Finally, a stable quantitative correlation between mouse weight and multidimensional features was established to ensure that the model has accurate weight mapping ability.

[0030] S6 calculates mouse weight parameters based on the association mapping model and outputs the mouse weight measurement results.

[0031] Specifically, step S6 calculates mouse weight parameters based on the association mapping model and outputs the mouse weight measurement results. The 48×30×10 dimensional core feature data selected in step S5 is standardized according to the input format required by the model, ensuring that the data values ​​are within the range of 0 to 1, avoiding the impact of differences in data magnitude on the calculation results. The standardized feature data is then input into the trained association mapping model, passed through the input layer to each hidden layer. The hidden layers progressively transform and fuse the feature data, extracting the key information most closely related to weight, and finally outputting the weight calculation result in a single numerical form through the output layer. During the calculation process, the model has a built-in result verification mechanism to judge the reasonableness of the output weight data. The valid range of the weight calculation results is set based on the common weight range of experimental mouse breeds. Data exceeding this range will trigger the model to recalculate, ensuring the reliability of the output results. After the calculation is completed, the weight measurement results are stored in a local database through the data output interface and simultaneously displayed in a visual form on the monitoring terminal, including single weight values ​​and continuous monitoring curves, facilitating real-time viewing and data traceability for researchers, providing direct weight data support for related experimental research.

[0032] Preferably, the expression for the mouse dynamic phase shift calibration prediction model is: ,in, These are the calibrated phase shift characteristic parameters. These are the original phase shift characteristic parameters. This represents the characteristic value of mouse movement rate. The time interval between adjacent frames. These are the model weight coefficients. This is the phase shift deviation correction factor. The time decay factor, This represents the gradient value of the original phase shift characteristic parameters.

[0033] Specifically, the mouse dynamic phase shift calibration prediction model is used to accurately correct phase shift bias in mouse dynamic image features. Its parameter values ​​were determined through extensive experimental validation: weighting coefficients were set to 0.8 and 0.3, phase shift bias correction coefficients were set to 1.2 and 0.9, and the time decay factor was set to 0.6. This parameter combination maximizes the reduction of phase shift interference caused by mouse movement and image acquisition time differences. During implementation, the original phase shift feature parameters are first obtained. These parameters are extracted from multi-view image sequences acquired by the mouse 3D behavioral phenotypic analysis platform, directly reflecting the degree of phase shift caused by viewpoint switching and mouse movement. Simultaneously, mouse movement rate feature values ​​are extracted to quantify the intensity of mouse movement during the acquisition period. The time interval corresponds to the time difference between adjacent frames, used to correct the cumulative phase shift error in the time dimension. The gradient values ​​of the original phase shift feature parameters are calculated using the model's built-in gradient solving module, reflecting the rate of change of the phase shift parameters. The model performs a nonlinear transformation on the linear combination of the original phase shift feature parameters and motion rate feature values ​​using the arctangent function, combines the time interval with an exponential function to decay the time interval, and then uses weight coefficients to weight and fuse the two results to finally output the calibrated phase shift feature parameters. These parameters can effectively eliminate the influence of phase shift bias on subsequent volume estimation and weight mapping, ensuring the accuracy of the data input to the gradient-optimized volume estimation network.

[0034] Preferably, the expression for the gradient-optimized volume estimation network is: ,in, For the estimated mouse three-dimensional volume correlation data, For the corrected feature information components, The number of feature components, , These are the network weight parameters. For network bias terms, For activation function, For gradient optimization coefficients, This is the gradient operator.

[0035] Specifically, the gradient-optimized volume estimation network achieves accurate estimation of mouse 3D volume correlation data through multi-level parameter optimization. The number of feature components is set to 80 and 60 based on the mouse feature extraction dimensions. The network weight parameters are iteratively optimized during training, with values ​​ranging from 0.01 to 0.05, 0.03 to 0.08, and 0.02 to 0.06, respectively. The bias term is set to 0.1, the sigmoid function is used as the activation function, and the gradient optimization coefficient is set to 0.7. In implementation, the corrected feature information is first decomposed into multiple feature components, each corresponding to different dimensions of feature data such as mouse contour, posture, and movement trajectory. After receiving the feature components, the network input layer performs preliminary data transformation through a linear combination of weight parameters and bias terms. The activation function performs non-linear processing on the transformation result to enhance the expressive power of the feature data and achieve preliminary extraction of volume correlation information. Simultaneously, the gradient operator calculates the gradient value of the sum of squares of some feature components. This gradient value reflects the correlation strength between the feature components and the volume estimation result. The gradient optimization coefficient is used to balance the weights of the linear transformation result and the gradient optimization result. Finally, by summing the processing results of each feature component, the estimated three-dimensional volume correlation data of the mouse is output. This data includes volume-related information of each body part of the mouse, providing accurate volume basis data for subsequent spatiotemporal phase alignment optimization.

[0036] Preferably, the expression for the spatiotemporal phase alignment optimization algorithm is: ,in, This is the volume data after spatiotemporal alignment. For the first Weight of each time node For the first Each spatial perspective weight, For the first Phase angle at each time node, For the first A spatial perspective phase angle, The total number of spatiotemporal nodes. For the estimated three-dimensional volume correlation data of mice.

[0037] Specifically, the spatiotemporal phase alignment optimization algorithm addresses the misalignment of volume data at different spatiotemporal nodes. The total number of spatiotemporal nodes is determined to be 48 based on the number of cameras and the acquisition duration. The weights of both time nodes and spatial viewpoints are determined using the analytic hierarchy process (AHP), with values ​​ranging from 0.05 to 0.25 to ensure a balanced contribution of data from each spatiotemporal node. During implementation, the 3D volume correlation data output by the gradient-optimized volume estimation network is first labeled with spatiotemporal nodes. Each time node corresponds to the volume data of a specific frame image, and each spatial viewpoint corresponds to the volume data acquired by different cameras. The phase angles of the time nodes and spatial viewpoints are obtained through the algorithm's built-in phase calculation module, quantifying the phase shift in the temporal dimension and the viewpoint phase difference in the spatial dimension, respectively. The algorithm calculates the sum of the products of the time node weights, spatial viewpoint weights, and the cosine of the phase difference, then divides this sum by the sum of the squared products of the time node weights and the squared spatial viewpoint weights to obtain the spatiotemporal alignment coefficient. This coefficient is used to correct the phase deviation of volume data at different spatiotemporal nodes. The alignment coefficient is multiplied by the three-dimensional volume correlation data to output the spatiotemporally aligned volume data. This data achieves unified calibration of volume data at different times and from different perspectives, eliminates spatiotemporal misalignment caused by mouse movement and camera layout, and provides standardized data for constructing a weight correlation mapping model.

[0038] Preferably, the expression for constructing the association mapping model between mouse body weight and multidimensional features is: ,in, The results are the mouse body weight measurement results. The correlation coefficient, This is the volume data after spatiotemporal alignment. These are the calibrated phase shift characteristic parameters. The gradient values ​​for the volume estimation data. For the first The weight coefficients of each feature component, For the corrected feature information components, This represents the total number of characteristic components.

[0039] Specifically, a correlation mapping model between mouse body weight and multidimensional features establishes a quantitative relationship between calibration parameters, volume data, calibration results, and mouse body weight. The correlation coefficients were determined to be 0.6, 0.4, and 0.5 through sample training. The total number of feature components was set to 50 based on the core feature selection results, and the weight coefficients ranged from 0.03 to 0.07. During implementation, calibration parameters output by the mouse dynamic phase shift calibration prediction model were first collected; these parameters reflect the degree of correction for phase shift bias. Volume correlation data generated by the gradient-optimized volume estimation network and the gradient values ​​of this data directly characterize the mouse volume characteristics and rate of change. The calibration results from the spatiotemporal phase alignment optimization algorithm provide standardized volume data. Simultaneously, the corrected feature information components were extracted, and core features strongly correlated with body weight were selected. The model weights the spatiotemporally aligned volume data, the product of the calibrated phase shift feature parameters and the gradient values ​​of the volume estimation data, and the sum of the products of the feature components and their corresponding weight coefficients using the correlation coefficients. The sum of these three weights yields the mouse body weight measurement result. This model integrates key data from the entire data chain, fully explores the intrinsic relationship between features of various dimensions and weight, ensures the accuracy of weight calculation, and achieves efficient mapping from multi-source data to weight results.

[0040] Preferably, the association model for the feature information collected by the mouse 3D behavioral phenotype analysis platform is as follows: ,in, To collect comprehensive feature information, For the collection coefficient, For the first Image acquisition weights from each perspective, For the first Image feature matrix from each perspective For the first Feature compensation values ​​for each viewpoint This represents the total number of viewpoints collected.

[0041] Specifically, the association model expression for feature information collected by the mouse 3D behavioral phenotypic analysis platform is used to integrate feature information collected from multiple perspectives. The acquisition coefficient is set to 0.9, the total number of acquisition perspectives corresponds to the number of cameras configured on the platform being 8, and the image acquisition weights for each perspective are set according to the importance of the camera's shooting angle, with values ​​ranging from 0.1 to 0.2. The feature compensation value is set to 0.05. During implementation, each camera on the platform acts as an independent acquisition perspective, acquiring a sequence of dynamic mouse images from that perspective. The image processing module extracts the image feature matrix from that perspective, which includes multi-dimensional feature data such as mouse posture, contour, and movement trajectory. The feature compensation value is used to correct feature loss caused by blind spots in single-view photography. By linearly combining the image feature matrix of each perspective with the corresponding acquisition weights and then superimposing the feature compensation value, the feature processing result for that perspective is obtained. Subsequently, the feature processing results of all perspectives are multiplied, and the product result is scaled and adjusted using the acquisition coefficients to finally output the comprehensive feature information collected. This model achieves effective fusion of feature data from multiple perspectives, fully utilizing the acquisition advantages of each perspective and compensating for the information limitations of a single perspective, providing comprehensive and complete raw feature data for subsequent phase shift correction and volume estimation.

[0042] Preferably, step S3 includes the following sub-steps: S31, performing hierarchical division on the corrected feature information, dividing the feature information into multiple basic processing units according to feature dimension, data type, and correlation with volume; S32, inputting different basic processing units into the corresponding layers of the gradient optimization volume estimation network, performing dimension matching and format conversion on the feature data through the network input layer to make the data meet the network computation requirements; S33, activating the gradient transfer mechanism inside the network, performing nonlinear transformation on the data through the hidden layer, and performing deep extraction of feature information by combining the parameter interaction of adjacent layers; S34, integrating the processed feature information based on the parameter configuration of the output layer to generate three-dimensional volume correlation data that can directly reflect the volume characteristics of the mouse.

[0043] Specifically, step S3 involves hierarchical data processing of the gradient-optimized volume estimation network. S31 involves hierarchically dividing the corrected feature information based on three dimensions: feature dimension, data type, and correlation with volume. The feature dimensions include three core categories: pose, contour, and motion trajectory. Data types are categorized as numerical and coordinate. A volume correlation threshold of 0.6 is set; features above the threshold are assigned to core processing units, while those below are assigned to auxiliary processing units. This results in 16 basic processing units, each containing 8-12 feature subsets. S32 inputs each basic processing unit to the corresponding layer of the gradient-optimized volume estimation network. The input layer performs format conversion of the feature data through a data interface, uniformly converting the feature subsets to 32-bit floating-point data and adjusting the dimension to 64×64 to ensure data consistency with volume. The number of nodes in the network input layer is matched to meet the format requirements of network computation. In S33, the gradient propagation mechanism inside the network is activated. The hidden layer uses the ReLU activation function to perform nonlinear transformation on the data. The gradient propagation step size is set to 0.002. The parameter interaction frequency between adjacent layers is once every 10 frames of data. The feature information is deeply extracted through weight matrix updates to mine key features related to mouse volume. In S34, based on the parameter configuration of the output layer, the number of nodes in the output layer is set to 32. The processed feature information is weighted and integrated. The weight coefficients are optimized to a value range of 0.02-0.08 through network training. During the integration process, weights are assigned according to the importance of features to generate three-dimensional volume correlation data that can directly reflect the volume features of the mouse trunk, limbs, head and other parts. The data dimension is 32×30, providing accurate volume basis data for subsequent spatiotemporal calibration.

[0044] Preferably, step S4 includes the following sub-steps: S41, extracting the time dimension identifier and spatial perspective identifier from the three-dimensional volume association data, establishing a spatiotemporal information index table, and clarifying the spatiotemporal position corresponding to different data; S42, based on the core logic of the spatiotemporal phase alignment optimization algorithm, calculating the phase difference and deviation value of volume data under different spatiotemporal nodes, and determining the calibration data nodes that need to be adjusted; S43, performing phase compensation and position calibration on the calibration data nodes according to the alignment rules preset by the algorithm, so that data of different spatiotemporal dimensions achieve consistent association; S44, integrating and verifying all calibrated volume data, removing abnormal data that does not meet the spatiotemporal alignment standard, and retaining valid data for subsequent processing.

[0045] Specifically, step S4 involves the step-by-step implementation of the spatiotemporal phase alignment optimization algorithm calibration logic. Accurate data alignment is achieved through multi-dimensional parameter settings. S41 extracts the time dimension identifier and spatial viewpoint identifier from the three-dimensional volume correlation data. The time dimension identifier is in frames, with frame numbers ranging from 1 to 1800. The spatial viewpoint identifier corresponds to eight acquisition cameras, numbered 1 to 8. A spatiotemporal information index table is established based on these identifiers. The index table uses a two-dimensional array structure, with row indices corresponding to time frame numbers and column indices corresponding to spatial viewpoint numbers, clearly defining the spatiotemporal coordinates of each data point. S42, based on the core logic of the spatiotemporal phase alignment optimization algorithm, sets the phase difference calculation window size to 5 frames. The phase difference of volume data at different spatiotemporal nodes is calculated using a sliding window. The deviation value is calculated using the mean square error method, with a deviation threshold set to 0. S45: Select key data nodes whose deviation values ​​exceed the threshold, and control the proportion of such nodes to within 15% of the total data nodes; S43: Perform phase compensation on key data nodes according to the alignment rules preset by the algorithm. The compensation coefficient is set to 0.1-0.3 according to the spatiotemporal position difference. At the same time, position calibration is performed with a calibration accuracy of 0.01 pixels to achieve consistent correlation between data in different spatiotemporal dimensions in terms of phase and position; S44: Integrate and verify all calibrated volume data. Use the threshold judgment method to set the verification threshold to 0.03, and remove abnormal data that deviate from the calibration benchmark by more than the threshold. The proportion of abnormal data removed does not exceed 5%. Valid data that meets the spatiotemporal alignment standard is retained. The spatiotemporal consistency error of valid data is controlled within 3%, providing standardized data support for the subsequent construction of the correlation mapping model.

[0046] Preferably, step S5 includes the following sub-steps: S51, collecting the calibration parameters of the mouse dynamic phase shift calibration prediction model, the volume correlation data of the gradient-optimized volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm to establish a multidimensional dataset; S52, performing feature screening on the multidimensional dataset, retaining core influencing features and removing redundant data based on the correlation strength between the data and mouse weight; S53, constructing the basic framework of the mapping model, setting the structure of the input layer, hidden layer, and output layer of the model, and clarifying the parameter transmission path of different levels; S54, inputting the screened core feature data into the model framework, adjusting the model parameters through data training, and determining the quantitative correlation between mouse weight and multidimensional features.

[0047] Specifically, step S5 involves constructing the correlation mapping model. S51 collects the calibration parameters of the mouse dynamic phase shift calibration prediction model, the volume correlation data of the gradient-optimized volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm. The calibration parameters include 64-dimensional bias correction coefficients. The volume correlation data is a 32×30 three-dimensional array, and the calibration results are 32×30×8-dimensional data after spatiotemporal alignment. The three types of data are concatenated along the time dimension to establish a multidimensional dataset with dimensions of 64×30×10. The data storage format uses binary encoding to improve processing efficiency. S52 performs feature screening on the multidimensional dataset. The Pearson correlation coefficient method is used to calculate the correlation strength between each feature and mouse body weight, setting a correlation threshold of 0.7. Core features with correlation strengths higher than the threshold are retained, with the core feature dimensions controlled to 48. Redundant data is removed, with the proportion of redundant data not exceeding 20%. S53 ensures the validity of the dataset; S53 constructs the basic framework of the mapping model, adopting a three-layer neural network structure. The number of nodes in the input layer is set to 48×30×10=14400, the hidden layer includes two layers with 1024 and 512 nodes respectively, and the output layer has 1 node. The layers are connected by a fully connected method, and the parameter transmission path is defined as input layer → first hidden layer → second hidden layer → output layer; S54 inputs the selected core feature data into the model framework, sets the training batch size to 32, the number of iterations to 800, the initial learning rate to 0.001, and decays by 50% every 200 iterations. The model parameters are adjusted through the backpropagation algorithm, and the parameter update step size is controlled between 0.0001 and 0.001, so that the error between the model prediction value and the measured value gradually converges to within 5%. Finally, the quantitative correlation between mouse weight and multidimensional features is determined, ensuring that the model has accurate weight mapping ability.

[0048] The mouse dynamic phase shift calibration prediction model is a calibration model built on a deep neural network to eliminate phase shift bias caused by motion and viewpoint switching during mouse dynamic image acquisition. The model consists of an input layer, four hidden layers, and an output layer. The input layer has 128 nodes, the hidden layers have 256, 512, 512, and 256 nodes respectively, and the output layer has 64 nodes. The model was trained using 3000 training samples labeled with phase shift bias, with 1000 training iterations and an initial learning rate of 0.001, decreasing by 50% every 200 iterations. The implementation process is as follows: First, a 128×30×60 dimensional feature matrix extracted from a mouse 3D behavioral phenotypic analysis platform is received, including 12 key features such as posture, contour, and motion trajectory. This matrix is ​​passed from the input layer to the hidden layers, undergoes a nonlinear transformation via an activation function, and uses the learned dynamic mapping relationship to adjust the phase shift-related bias data. The correction amount is controlled between -0.15 and 0.15, and finally, 64-dimensional calibrated phase shift feature parameters are output. This model corrects phase shift errors during image acquisition, avoiding distortion of feature data caused by changes in mouse movement speed and differences in camera perspective. It provides accurate raw data support for subsequent volume estimation, solves the problem that traditional methods struggle to handle dynamic phase shift bias, ensures the accuracy of weight measurement from the data source, and lays the foundation for the reliability of the entire technical solution.

[0049] The Gradient-Optimized Volume Estimation Network is an intelligent network with hierarchical data processing capabilities, used to extract and generate 3D volume correlation data for mice from corrected feature information. The network comprises three main modules: feature enhancement, gradient calculation, and volume correlation data generation. The feature enhancement module has six convolutional layers with alternating kernel sizes of 3×3 and 5×5, all with a stride of 1. The gradient calculation module uses the backpropagation algorithm with a gradient threshold of 0.005. The volume generation module contains two fully connected layers with 128 and 64 nodes respectively. Its implementation process is as follows: First, corrected feature information is received and enhanced through convolution and pooling in the feature enhancement module. Then, the gradient calculation module calculates the gradient values, updating the network parameters every 50 frames with a stride of 0.001 to 0.003. Finally, the fully connected layers integrate the processing results to generate 64×30 dimensional volume correlation data, including volume information for the trunk, limbs, and head. This network accurately mines the intrinsic correlation between feature information and mouse volume through gradient optimization and hierarchical parameter iteration, generating quantitative volume data. It breaks through the accuracy limitations of traditional volume estimation methods. Through intelligent feature extraction and parameter optimization, it achieves efficient and accurate estimation of mouse volume, providing core data support for weight correlation mapping models and promoting the transformation of non-contact weight measurement from qualitative analysis to quantitative calculation.

[0050] The spatiotemporal phase alignment optimization algorithm is used to achieve unified calibration of volumetric data from different spatiotemporal nodes, specifically addressing data misalignment issues in multi-view, dynamic acquisition scenarios. The algorithm sets a 10-frame time calibration window, assigns weights of 0.2 to frontal and side cameras and 0.1 to oblique cameras for the eight acquisition perspectives, sets a total of 48 spatiotemporal nodes, a phase difference control threshold of 5°, and a calibration precision of 0.01 pixels. The implementation process is as follows: First, extract the time frame identifiers (frames 1-1800) and spatial viewpoint identifiers (cameras 1-8) from the 3D volumetric correlation data, establishing a two-dimensional array-structured spatiotemporal index table; then, calculate the phase difference using a 5-frame sliding window, employing the mean square error method to filter key data nodes with deviations exceeding 0.05 (≤15%); subsequently, perform phase compensation with a compensation coefficient of 0.1-0.3, combined with position calibration at a precision of 0.01 pixels; finally, remove abnormal data using a verification threshold of 0.03 (≤5%), generating 64×30×8 dimensional spatiotemporally aligned volumetric data. This algorithm eliminates phase accumulation error in the time dimension and perspective deviation in the spatial dimension, establishes a unified standard volume data system, solves the data fragmentation problem caused by multi-view dynamic acquisition, and achieves consistent correlation of volume data under different scenarios through precise calibration in the spatiotemporal dimension. It provides standardized input data for the weight mapping model and ensures the stability and reliability of weight measurement results under different experimental conditions.

[0051] The mouse 3D behavioral phenotypic analysis platform is a hardware and software integrated system that combines multi-view image acquisition and feature extraction. The platform is equipped with eight evenly distributed high-definition industrial cameras with a resolution of 3840×2160 pixels, a frame rate of 60 frames per second, an acquisition time of 30 seconds, and a synchronization trigger error of ≤10 milliseconds. Its implementation involves simultaneously acquiring dynamic image sequences of mice through the eight cameras, covering various behavioral states such as stillness, movement, and turning. The acquired image data is transmitted to a local server in real time. Subsequently, through built-in image segmentation and feature extraction algorithms, 12 types of pose features, coordinates of 100 contour feature points, and information such as the center of gravity trajectory are extracted, ultimately forming a 128×30×60-dimensional feature matrix. This platform provides comprehensive and high-quality raw data for subsequent algorithms and models. It ensures the integrity of feature information through multi-view acquisition and guarantees the accuracy of feature data through high-precision image extraction. It breaks through the information limitations of traditional single-view acquisition and realizes the all-round and dynamic capture of mouse behavioral characteristics. It provides sufficient data support for subsequent steps such as phase shift correction, volume estimation, and spatiotemporal alignment. It is the basic guarantee for the entire contactless weight measurement solution and provides reliable hardware and data acquisition support for the application of artificial intelligence technology in weight measurement.

[0052] like Figure 2As shown, an artificial intelligence-based method for determining mouse weight is implemented through different units, including: an image sequence acquisition and feature extraction unit, used to acquire dynamic image sequences of mice from different perspectives through a preset image acquisition device, separate and extract feature information representing the mouse's posture, contour, and movement trajectory from the images, and transmit the extracted feature information to a dynamic phase shift deviation correction unit; a dynamic phase shift deviation correction unit, which receives the feature information transmitted by the feature extraction unit, calls a mouse dynamic phase shift calibration prediction model to correct the phase shift deviation in the feature information, and sends the corrected feature information to a three-dimensional volume association data generation unit; and a three-dimensional volume association data generation unit, which receives the corrected feature information, processes the data through a gradient-optimized volume estimation network, and generates a three-dimensional volume association data for the mouse. The system receives 3D volume correlation data and transmits it to the spatiotemporal dimension calibration unit. The spatiotemporal dimension calibration unit receives the 3D volume correlation data, performs spatiotemporal dimension calibration using a spatiotemporal phase alignment optimization algorithm, establishes alignment mappings for volume data at different spatiotemporal nodes, and transmits the calibrated data to the weight correlation mapping model construction unit. The weight correlation mapping model construction unit collects the calibration parameters, volume correlation data, and calibration results output by different units, constructs a correlation mapping model between mouse weight and multidimensional features, and transmits the model to the weight parameter calculation unit. The weight parameter calculation unit receives the correlation mapping model, calculates the mouse weight parameters based on the quantitative correlation relationships set in the model, and outputs the mouse weight measurement results. Different units interact and collaborate bidirectionally through the data transmission interface.

[0053] The AI-based mouse weight measurement method, through specially designed dynamic phase shift calibration, gradient-optimized volume estimation, and spatiotemporal phase alignment related technologies, forms a complete processing flow from multi-view feature acquisition to weight mapping. It can effectively capture key information such as posture changes and trajectory deviations in the dynamic behavior of mice, accurately correct phase shift bias and spatiotemporal data misalignment, and solve the shortcomings of existing technologies that lack dedicated optimization mechanisms, resulting in insufficient measurement accuracy. At the same time, through the deep integration of three-dimensional behavioral phenotypic analysis and weight correlation model, it achieves accurate adaptation to the physiological characteristics and behavioral patterns of mice, and significantly improves the stability of weight measurement in different scenarios.

[0054] This method, through the automated data processing capabilities of an intelligent model, achieves a fully automated process from image acquisition and feature extraction to weight calculation. This avoids stress responses in mice caused by manual weighing, ensuring the objectivity of experimental data, and significantly improves the efficiency of weight monitoring. It meets the needs of high-frequency, batch measurements in large-scale experiments, overcoming the shortcomings of existing technologies, such as low efficiency and difficulty in adapting to large-scale research. At the same time, the collaborative work of various technical modules and the optimized design of the entire chain ensure that every link from data acquisition to result output accurately serves the goal of weight measurement, further enhancing the practicality and reliability of the method and providing more efficient and accurate technical support for research in related fields.

[0055] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for determining mouse body weight based on artificial intelligence, characterized in that, Includes the following steps: S1. A multi-view dynamic image sequence of mice was collected through a mouse 3D behavioral phenotyping platform, and feature information representing mouse posture, contour and movement trajectory was extracted from the images. S2, call the mouse dynamic phase shift calibration prediction model to correct the phase shift deviation of the collected feature information, and adjust the deviation data related to phase shift in the feature information through the dynamic mapping relationship constructed by the model; S3 uses a gradient-optimized volume estimation network to process the corrected feature information. Through gradient transfer and parameter iterative update between network layers, mouse three-dimensional volume correlation data is generated. S4. The spatiotemporal phase alignment optimization algorithm is used to perform spatiotemporal dimension calibration on the three-dimensional volume correlation data, and establish the alignment mapping of volume data under different time nodes and spatial perspectives. S5. Combining the correction parameters output by the mouse dynamic phase shift calibration prediction model, the volume correlation data generated by the gradient optimization volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm, a correlation mapping model between mouse body weight and multidimensional features is constructed. S6 calculates mouse weight parameters based on the association mapping model and outputs the mouse weight measurement results.

2. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, The expression for the mouse dynamic phase shift calibration prediction model is: ,in, These are the calibrated phase shift characteristic parameters. These are the original phase shift characteristic parameters. This represents the characteristic value of mouse movement rate. The time interval between adjacent frames. These are the model weight coefficients. This is the phase shift deviation correction factor. The time decay factor, This represents the gradient value of the original phase shift characteristic parameters.

3. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, The expression for the gradient-optimized volume estimation network is: ,in, For the estimated mouse three-dimensional volume correlation data, For the corrected feature information components, The number of feature components, , These are the network weight parameters. For network bias terms, For activation function, For gradient optimization coefficients, This is the gradient operator.

4. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, The expression for the spatiotemporal phase alignment optimization algorithm is: ,in, This is the volume data after spatiotemporal alignment. For the first Weight of each time node For the first Each spatial perspective weight, For the first Phase angle at each time node, For the first A spatial perspective phase angle, The total number of spatiotemporal nodes. For the estimated three-dimensional volume correlation data of mice.

5. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, The expression for constructing the association mapping model between mouse body weight and multidimensional features is as follows: ,in, The results are the mouse body weight measurement results. The correlation coefficient, This is the volume data after spatiotemporal alignment. These are the calibrated phase shift characteristic parameters. The gradient values ​​for the volume estimation data. For the first The weight coefficients of each feature component, For the corrected feature information components, This represents the total number of characteristic components.

6. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, The association model for the feature information collected by the mouse 3D behavioral phenotype analysis platform is as follows: ,in, To collect comprehensive feature information, For the collection coefficient, For the first Image acquisition weights from each perspective, For the first Image feature matrix from each perspective For the first Feature compensation values ​​for each viewpoint This represents the total number of viewpoints collected.

7. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, S3 includes the following steps: S31, The corrected feature information is hierarchically divided into multiple basic processing units according to feature dimension, data type and correlation with volume. S32, different basic processing units are input into the corresponding layers of the gradient optimization volume estimation network. The feature data is dimension-matched and format-converted through the network input layer so that the data meets the network's computational requirements. S33, initiates the gradient transfer mechanism within the network, performs nonlinear transformation on the data through the hidden layer, and performs deep feature extraction by combining the parameter interaction of adjacent layers; S34 integrates the processed feature information based on the parameter configuration of the output layer to generate three-dimensional volume correlation data that can directly reflect the volume characteristics of the mouse.

8. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, S4 includes the following steps: S41, extract the time dimension identifier and spatial perspective identifier from the three-dimensional volume correlation data, establish a spatiotemporal information index table, and clarify the spatiotemporal location corresponding to different data; S42, based on the core logic of the spatiotemporal phase alignment optimization algorithm, calculates the phase difference and deviation value of volume data under different spatiotemporal nodes, and determines the calibration data nodes that need to be adjusted; S43, according to the alignment rules preset by the algorithm, performs phase compensation and position calibration on the calibrated data nodes to achieve consistent correlation of data in different spatiotemporal dimensions; S44 integrates and verifies all calibrated volume data, removes abnormal data that does not conform to the spatiotemporal alignment standard, and retains valid data for subsequent processing.

9. The method for determining mouse weight based on artificial intelligence according to claim 1, characterized in that, S5 includes the following steps: S51, collect the calibration parameters of the mouse dynamic phase shift calibration prediction model, the volume correlation data of the gradient-optimized volume estimation network, and the calibration results of the spatiotemporal phase alignment optimization algorithm to establish a multidimensional dataset; S52 performs feature filtering on the multidimensional dataset, retaining core influencing features and removing redundant data based on the strength of the correlation between the data and mouse body weight. S53, Construct the basic framework of the mapping model, define the structure of the input layer, hidden layer and output layer of the model, and clarify the parameter transmission path of different layers; S54. Input the selected core feature data into the model framework, adjust the model parameters through data training, and determine the quantitative correlation between mouse weight and multidimensional features.

10. The method for determining mouse weight based on artificial intelligence according to any one of claims 1-9, characterized in that, This method is implemented through different units, including: The image sequence acquisition and feature extraction unit is used to acquire dynamic image sequences of mice from different perspectives through a preset image acquisition device, separate and extract feature information representing the mouse's posture, contour and movement trajectory from the images, and transmit the extracted feature information to the dynamic phase shift deviation correction unit. The dynamic phase shift deviation correction unit receives the feature information transmitted by the feature extraction unit, calls the mouse dynamic phase shift calibration prediction model to correct the phase shift deviation in the feature information, and sends the corrected feature information to the three-dimensional volume correlation data generation unit. The three-dimensional volume correlation data generation unit receives the corrected feature information, processes the data through a gradient optimization volume estimation network, generates mouse three-dimensional volume correlation data, and transmits it to the spatiotemporal dimension calibration unit. The spatiotemporal dimension calibration unit receives three-dimensional volume correlation data, uses a spatiotemporal phase alignment optimization algorithm to perform spatiotemporal dimension calibration, establishes alignment mapping of volume data under different spatiotemporal nodes, and transmits the calibrated data to the weight correlation mapping model construction unit. The weight-related mapping model construction unit collects the correction parameters, volume-related data and calibration results output by different units, constructs the correlation mapping model between mouse weight and multidimensional features, and transmits the model to the weight parameter calculation unit. The weight parameter calculation unit receives the correlation mapping model, calculates the mouse weight parameters based on the quantitative correlation relationship set in the model, and outputs the mouse weight measurement results. Different units can interact and work together through a data transmission interface.

Images