A pet travel whole-process management method and system based on multi-source perception and real-time interaction

Abstract

The present application relates to the technical field of data processing, in particular to a pet travel whole-process management method and system based on multi-source sensing and real-time interaction, comprising the following steps: removing out-of-bound difference through collecting multi-dimensional physiological and environmental data to generate a synchronous data set, weighting and adding the data to generate a physiological evaluation result, extracting video brightness variation pixels to generate a behavior feature vector, fusing the features to input a classification model to generate a state evaluation result, comparing rules to extract matching parameters to generate a travel management result.In the present application, by fusing multi-dimensional physiological and environmental data, combining image brightness variation to extract behavior features, and through model space mapping to obtain a state evaluation result, automatic comparison is performed to output ventilation adjustment and interaction parameters, information blind spots caused by manual determination are overcome, dynamic constraints and abnormal intervention are realized, and pet health protection and response agility are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for managing the entire pet travel process based on multi-source perception and real-time interaction. Background Technology

[0002] The field of data processing technology refers to the technical system that uses computer systems to collect, transmit, store, analyze and manage various types of data. It covers core aspects such as information acquisition, data organization, business process processing and cross-terminal interaction, and is widely used in scenarios such as transportation, logistics scheduling and service management. It achieves process control and information collaboration by integrating multi-source data and combining it with business rules.

[0003] The traditional pet travel management method and system refers to the multi-stage management of pets during transportation, pick-up and drop-off and care. It involves handling specific matters such as reservation, scheduling, transportation and delivery during the pet travel process by manually registering pet information, placing transportation orders by phone or application, arranging vehicles for pick-up and drop-off according to fixed routes, and having staff record the pet's status at multiple points and summarize the information after transportation is completed.

[0004] The existing traditional pet travel management system relies on staff to record status at key points and summarize the data afterward. This operating model is prone to creating information blind spots during transportation. Because it relies solely on manual observation and periodic reporting, it is impossible to detect sudden stress reactions and physiological abnormalities in pets in real time. When the environment inside the crate changes drastically, the delayed information acquisition mechanism makes it difficult to intervene in abnormal situations in a timely manner, greatly increasing the risk of damage to the pet's health. At the same time, subjective judgment has limitations such as inconsistent standards and susceptibility to interference, resulting in a lack of scientific and dynamic supervision and constraints on the overall transportation process, which directly restricts the safety and response efficiency of the travel management process. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention provides a method for managing the entire pet travel process based on multi-source perception and real-time interaction, comprising the following steps: S1: Collect pet heartbeat electrical signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, combine them with preset time span thresholds to remove out-of-bounds time differences, and generate a synchronous dataset; S2: Weight the synchronous dataset by multiplying the electrocardiogram signal, body surface temperature and voltage, chamber temperature and voltage, humidity and oxygen concentration by their respective weight coefficients and summing them to generate physiological evaluation results. S3: Obtain the video image inside the transport box at the time corresponding to the physiological evaluation result and perform adaptive frame extraction in combination with the physiological evaluation score. Subtract the brightness of the pixels at the corresponding coordinate positions of adjacent frames and extract the pixels with brightness changes to generate a behavior feature vector. S4: The physiological evaluation results are concatenated and aggregated with the behavioral feature vectors and input into a support vector machine for classification. The feature fusion results are mapped and classified based on the spatial partitioning hyperplane to generate state evaluation results. S5: Compare the status assessment results with the preset hierarchical rule configuration table, extract the environmental ventilation speed parameters and interactive prompt codes that match the status assessment results based on the association mapping logic, and encapsulate them to generate travel management results.

[0006] As a further aspect of the present invention, the synchronized dataset includes a heart rate electrical signal sequence, a body surface temperature and voltage sequence, an oxygen concentration change sequence, and a humidity change sequence; the physiological evaluation results include a physiological score, a physiological stability index, and a multi-parameter fusion index; the behavioral feature vector includes a brightness change distribution, a motor activity feature, and a pixel difference statistic; the state assessment results include a state category label, a state confidence score, and a classification boundary mapping parameter; and the travel management results include ventilation speed adjustment parameters, an environmental control instruction set, and interactive prompt codes.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Collects pet heartbeat electrical signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, timestamps the output values ​​of multiple sensor channels in chronological order, and aligns the values ​​at the same time index position to obtain multi-source time-stamped data; S102: Based on the multi-source time-labeled data, call the multi-channel timestamp sequence, calculate the difference between adjacent timestamps, compare the obtained difference with the preset time span threshold item by item, index and mark the positions where the difference exceeds the time span threshold and remove the corresponding data rows to obtain the time difference constraint filtering dataset; S103: Based on the time difference constraint, filter the dataset, call the remaining data row index set, perform unified time series rearrangement on the multi-channel values, and perform index compression on the missing positions to obtain the synchronized dataset.

[0008] As a further aspect of the present invention, the time span threshold is analyzed by extracting the hardware reference sampling period parameter of the multi-source sensor and the fixed delay time constant of the communication link, extracting the first and last timestamp span values ​​of the initial sample sequence and dividing them by the total number of samples to calculate the mean compensation amount, and combining the mean compensation amount with the sum of the reference sampling period parameter and the fixed delay time constant to determine the value.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the multi-data input hierarchical analysis method of the synchronous dataset, weight allocation is performed, a pairwise comparison sequence is constructed for multiple data dimensions and the feature vector is solved, the multiple components of the feature vector are normalized and data components that do not meet the preset consistency ratio threshold are filtered out, and feature weight coefficient vector is generated. S202: Based on the feature weight coefficient vector, call the synchronous dataset of heartbeat signal, body surface temperature and voltage, chamber temperature and voltage, humidity and oxygen concentration data, multiply the values ​​of multiple data channels with the corresponding weight components and accumulate them one by one to obtain the multi-source weighted fusion result. S203: Based on the multi-source weighted fusion result, call the data channel identifier corresponding to the multi-weighted item and perform linear normalization mapping, match the normalization result with the preset evaluation interval boundary value and perform segment calibration, combine the calibration segment index with the fusion value and encode to generate physiological evaluation result.

[0010] As a further aspect of the present invention, the consistency ratio threshold is determined by calling the matched average random consistency index through the order constant of the pairwise comparison sequence, and performing multiplicative scaling calculation in combination with the judgment fault tolerance constant. The numerical limit obtained by the scaling calculation is then summed and compensated with the underlying data scaling factor to determine the final value.

[0011] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the physiological evaluation results, obtain the corresponding time series and perform frame extraction analysis on the video images inside the pet transport box. Extract continuous frame sequences and align the timestamps of multiple frames. Match, verify, and interpolate the frame timestamps with the time series of the physiological evaluation results to obtain a time-synchronized frame sequence index set. S302: According to the time synchronization frame sequence index set, call the corresponding frame image data, traverse the pixel data of each frame, extract the brightness of multiple pixels, and aggregate the brightness of all pixels under the same timestamp by spatial coordinate mapping, while performing sequential encoding to generate a pixel brightness sequence. S303: Perform brightness value difference operation on the corresponding coordinate pixels of adjacent frames for the pixel brightness sequence, encode the multiple pixels in the difference result in time order and perform vectorization mapping to generate behavioral feature vectors.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Align the physiological evaluation results with the behavioral feature vector according to the corresponding time points, perform index matching based on the timestamp sequence, perform column-wise concatenation operation on the matched numerical vector according to the field order, and perform interval normalization on the concatenated multi-column values ​​to obtain fused feature data. S402: Based on the fused feature data, perform classification interval measurement calculation on multiple sample vectors, group and map the sample vectors according to the category labels, calculate the support vector candidate set for multiple groups of sample vectors, and filter the vector set that meets the interval constraint to establish the support vector set. S403: Based on the support vector set, perform spatial partitioning operation on the multi-sample vectors in the fused feature data, construct hyperplane parameters, perform symbol mapping calculation on the multi-sample vectors, match the mapping results with the category codes, and generate state evaluation results.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: The state assessment result is compared with the preset hierarchical rule configuration table. Multiple state indicators in the state assessment result are compared with the corresponding level ranges in the hierarchical rule configuration table. The rule index identifier is matched and marked according to the range of the indicator, and a state hierarchical index sequence is generated. S502: Based on the state classification index sequence, call the associated fields in the classification rule configuration table, extract the ventilation speed corresponding to each index identifier and perform normalization conversion, map the speeds corresponding to multiple indices to a unified scale sequence, and at the same time associate the encoding values ​​of the interactive prompt field to generate a ventilation speed parameter encoding set. S503: Based on the ventilation speed parameter encoding set, the speed parameter sequence and interactive prompt encoding are concatenated, the check bit is calculated and the frame header identifier is filled according to the data frame structure order, and the concatenated data sequence is integrated with binary encoding to generate travel management results.

[0014] A pet travel end-to-end management system based on multi-source sensing and real-time interaction includes: The data analysis module collects pet heart rate signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, combines preset time span thresholds to remove out-of-bounds time differences, generates a synchronous dataset and transmits it to the physiological assessment module. The physiological assessment module assigns weights to the synchronous dataset, multiplies and sums the heart rate electrocardiogram, body surface temperature and voltage, chamber temperature and voltage, humidity, and oxygen concentration with their corresponding weight coefficients, generates physiological assessment results, and transmits them to the behavior extraction module. The behavior extraction module acquires the video image inside the transport box at the time corresponding to the physiological evaluation result and performs adaptive frame extraction in combination with the physiological evaluation score. It subtracts the brightness of the pixels at the corresponding coordinate positions of adjacent frames and extracts the pixels with brightness changes, generates a behavior feature vector, and passes it to the state classification module. The state classification module concatenates and aggregates the physiological evaluation results with the behavioral feature vectors and inputs them into a support vector machine for classification. It performs mapping and classification calculations on the feature fusion results based on the spatial partitioning hyperplane, generates state evaluation results, and transmits them to the travel management module. The travel management module compares the status assessment results with the preset hierarchical rule configuration table, extracts the environmental ventilation speed parameters and interactive prompt codes that match the status assessment results based on the association mapping logic, and encapsulates them to generate travel management results.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, multi-dimensional objective environmental and physiological data are collected synchronously and feature weights are assigned. Behavioral feature vectors are constructed by combining image brightness difference analysis. The model is used to spatially map and classify multi-source fusion features to obtain state assessment results. Based on rule comparison, ventilation adjustment parameters and interaction codes are automatically output. This effectively overcomes the information blind spots caused by inconsistent subjective judgment standards and periodic reporting, transforms lagging supervision into automated real-time dynamic constraints, and ensures that environmental intervention can be triggered immediately when abnormal situations occur. This comprehensively enhances the level of pet health protection and business response agility in transportation scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] Please see Figure 1 This invention provides a method for managing the entire pet travel process based on multi-source perception and real-time interaction, including the following steps: S1: Collect pet heartbeat electrical signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, combine them with preset time span thresholds to remove out-of-bounds time differences, and generate a synchronous dataset; S2: Weight the synchronous dataset by multiplying the electrocardiogram signal, body surface temperature and voltage, chamber temperature and voltage, humidity, and oxygen concentration by their respective weight coefficients and summing them to generate physiological evaluation results. S3: Obtain video images inside the transport box at the time corresponding to the physiological evaluation results and perform adaptive frame extraction in combination with the physiological evaluation scores. Subtract the brightness of the pixels at the corresponding coordinate positions of adjacent frames and extract the pixels with brightness changes to generate a behavioral feature vector. S4: The physiological evaluation results are concatenated and aggregated with the behavioral feature vectors and input into the support vector machine for classification. The feature fusion results are mapped and classified based on the spatial partitioning hyperplane to generate the state evaluation results. S5: Compare the status assessment results with the preset hierarchical rule configuration table, extract the environmental ventilation speed parameters and interactive prompt codes that match the status assessment results based on the association mapping logic, and encapsulate them to generate travel management results.

[0021] The synchronous dataset includes cardiac electrocardiogram sequences, body surface temperature and voltage sequences, oxygen concentration change sequences, and humidity change sequences. Physiological evaluation results include physiological scores, physiological stability indices, and multi-parameter fusion indices. Behavioral feature vectors include brightness change distribution, motor activity features, and pixel difference statistics. State assessment results include state category labels, state confidence, and classification boundary mapping parameters. Travel management results include ventilation speed adjustment parameters, environmental control instruction sets, and interactive prompt codes.

[0022] Please see Figure 2 The specific steps of S1 are as follows: S101: Collects pet heartbeat electrical signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, timestamps the output values ​​of multiple sensor channels in chronological order, and aligns the values ​​at the same time index position to obtain multi-source time-stamped data; The system acquires analog values ​​of pet heartbeat signals from an attached ECG sensor, analog values ​​of body surface temperature and voltage continuously collected by an infrared temperature sensor, analog values ​​of internal temperature and voltage output from a thermistor sensor placed inside the enclosure, and oxygen concentration percentage data measured by an electrochemical oxygen sensor and humidity percentage data fed back by a capacitive humidity sensor. Analog-to-digital conversion is performed on the raw analog signals acquired from these five sensor channels, generating a digital signal sequence with uniform dimensions through a conversion channel with a sampling frequency of 1000 Hz. A moving average filtering algorithm is then used to denoise the generated digital signal sequence. Specifically, the sliding window contains 10 numerical nodes, the arithmetic mean and statistical standard deviation of the 10 numerical nodes are calculated, and outlier noise data exceeding the range of the arithmetic mean plus or minus twice the statistical standard deviation are removed. A timestamp is assigned to the denoised digital signal sequence of each sensor channel, with a minimum time resolution of 1 millisecond, and the recording operation is performed strictly in the order in which the physical signals were acquired. The system acquires timestamp sequences generated between different sensor channels, extracts the timestamp sequence of the temperature and voltage inside the enclosure as the baseline time axis sequence, and iterates through the timestamp nodes of the other four sensor channels. It determines the absolute value of the time deviation between the timestamps of other channels and a certain time node in the baseline time axis sequence. If the absolute value of this time deviation is less than 5 milliseconds, the time node is determined to be at the same time index position. The monitoring values ​​of each channel belonging to the same time index position are horizontally combined and spliced ​​to form a single-line data record containing a unified timestamp and five monitoring values. For example, when the baseline timestamp is 16000 milliseconds, the corresponding collected temperature and voltage inside the enclosure is extracted as 2.5 volts, the pet's heartbeat signal amplitude is 1.2 millivolts, the body surface temperature and voltage is 1.8 volts, the oxygen concentration percentage is 21.5%, and the humidity percentage is 45.0%. These five dimensions, along with the timestamp, are arranged in one row. The above matching and splicing alignment operations are continuously performed to generate multi-source time-labeled data containing multiple rows of continuous data records.

[0023] S102: Based on multi-source time-labeled data, call multi-channel timestamp sequences, calculate the difference between adjacent timestamps, compare the obtained difference with the preset time span threshold item by item, index and mark the positions where the difference exceeds the time span threshold and remove the corresponding data rows to obtain the time difference constraint filtering dataset; Based on the acquired multi-source time-labeled data, the timestamp sequence corresponding to the baseline time axis is extracted. An adjacent timestamp difference calculation is performed on this timestamp sequence. Specifically, the current timestamp value of the current data row in the timestamp sequence is extracted, and the preceding timestamp value of the previous adjacent data row is obtained. The difference between the current timestamp value and the preceding timestamp value is calculated, and this absolute time difference is used as the adjacent time interval. 1000 consecutive valid time interval records collected during the historical normal operation cycle are acquired. The arithmetic mean of these 1000 time interval values ​​is calculated and used as the baseline time span. The statistical standard deviation of these 1000 time intervals is also calculated. The baseline time span is summed with three times the statistical standard deviation to derive a preset time span threshold. For example, if the baseline time span is calculated to be 10 milliseconds and the statistical standard deviation is 1 millisecond, summing 10 milliseconds with 3 multiplied by 1 millisecond yields a preset time span threshold of 13 milliseconds. Each calculated adjacent time interval is then compared with the preset time span threshold of 13 milliseconds. When a specific adjacent time interval is determined to be greater than a preset time span threshold of 13 milliseconds, the data row containing the current timestamp is marked as abnormal, and a forced removal operation is performed to completely delete the abnormal data row from the multi-source time-labeled data. For example, the timestamp of the valid data in row 50 is 16050 milliseconds, and the timestamp of the data to be tested in row 51 is 16065 milliseconds. The difference between 16065 milliseconds and 16050 milliseconds is 15 milliseconds. Since this 15 milliseconds is greater than the preset time span threshold of 13 milliseconds, the data in row 51 is marked as abnormal and the entire row is removed. After performing the above filtering on all data rows, the remaining data rows that have not been removed together constitute the time difference constraint filtering dataset. Table 1 below is an example table of multi-source time-labeled data.

[0024] Table 1. Examples of Multi-Source Time Labeling

[0025] As shown in Table 1, by comparing and filtering the time interval with the preset time span threshold, rows carrying the retention flag are extracted to generate the final processed time difference constraint filtering dataset.

[0026] S103: Filter the dataset according to the time difference constraint, call the remaining data row index set, perform unified time series rearrangement on the multi-channel values, and perform index compression on the missing positions to obtain the synchronous dataset; The remaining data rows in the dataset are filtered using time difference constraints, and the inherent row numbers of all undone data rows are extracted to form a remaining data row index set. Based on this index set, the corresponding data matrix is ​​extracted. A unified time series rearrangement operation is performed on the multi-channel sensor values ​​in the dataset. The timestamp corresponding to the first row number in the remaining data row index set is extracted as the starting time reference point, and a standard time axis sequence with a fixed time step of 10 milliseconds is generated. The native timestamps carried in the remaining data rows are sequentially matched with each standard time node generated in the standard time axis sequence. If the absolute time deviation between a native timestamp and a standard time node is within a tolerance range of 2 milliseconds, the five multi-channel values ​​contained in that row are directly attached to the corresponding standard time node. For the blank time node positions left after the aforementioned culling operation, index compression is performed. The blank missing time nodes in the generated standard time axis sequence are traversed, and the specific number of consecutively existing missing nodes is counted. When the number of consecutive missing nodes is less than or equal to 3, the monitoring values ​​of each channel of the two adjacent valid data rows at the missing position are extracted in terms of time. The compensation value is derived by linearly weighting and summing according to the time span ratio, and then filled into the corresponding missing node position. When the number of consecutive missing nodes is greater than 3, the data segment is determined to be invalid. The consecutive blank standard time nodes of the segment are directly deleted. The timestamp values ​​of the first valid data row immediately following the blank node and all subsequent valid data rows are extracted. The timestamp values ​​of the above subsequent rows are shifted forward in a decreasing manner. The specific shift decrease is equal to the number of consecutive missing nodes multiplied by a time step of 10 milliseconds. The two consecutively missing valid data segments are then directly concatenated end to end on the time-series index sequence. For example, if the last time node of the preceding valid data is 16050 milliseconds, subsequent scanning reveals four consecutive missing nodes, creating a 40-millisecond time gap. The starting node of the subsequent valid data is 16100 milliseconds. A translation and compression operation is used to change 16100 milliseconds to 16060 milliseconds, and all timestamps after 16100 milliseconds are sequentially subtracted by 40 milliseconds. After the above weighted summation, padding, and decreasing translation and concatenation operations, the final normalized output synchronized dataset is obtained.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the multi-data input hierarchical analysis method of the synchronous dataset, weight allocation is performed, a pairwise comparison sequence is constructed for multiple data dimensions and the feature vector is solved, the multiple components of the feature vector are normalized and the data components that do not meet the preset consistency ratio threshold are filtered out, and the feature weight coefficient vector is generated. The synchronous dataset was extracted to include the following dimensions: electrocardiogram (ECG), body surface temperature and voltage, incubator temperature and voltage, oxygen concentration percentage, and humidity percentage. Pairwise comparison sequences were constructed for these five data dimensions. Prior importance evaluation indicators were retrieved from the historical pet physiological monitoring database to extract relative importance scores between dimensions. The relative importance score of the ECG dimension relative to the body surface temperature and voltage dimension was calculated to be 3, and the relative importance score of the ECG dimension relative to the incubator temperature and voltage dimension was calculated to be 5. Based on this extraction method, a multi-dimensional pairwise comparison numerical array was constructed. Feature vector extraction was performed on this pairwise comparison numerical array. The vertical distribution values ​​of each column in the numerical array were extracted and summed to obtain the sum of each vertical column. Each original value in the numerical array was divided by the sum of its corresponding vertical column to obtain an intermediate normalized array. The distribution values ​​of each horizontal row of this intermediate normalized array were extracted and the arithmetic mean was calculated. The five arithmetic means were arranged in the original dimension order to generate the initial feature vector components. The initial feature vector components for the cardiac electrocardiogram dimension are extracted as follows: 0.45 for the cardiac electrocardiogram, 0.25 for the body surface temperature and voltage dimension, 0.15 for the chamber temperature and voltage dimension, 0.10 for the oxygen concentration percentage dimension, and 0.05 for the humidity percentage dimension. A preset consistency ratio threshold is obtained, calculated by extracting 100 logically consistent historical pairwise comparison value arrays and determining their upper limit of consistency distribution; this threshold is set to 0.10. The maximum eigenvalue of the value array is extracted and combined with the dimension count of 5 to calculate the actual consistency ratio, which, after multiplication and division, yields an actual consistency ratio of 0.04. This 0.04 is compared with the preset consistency ratio threshold of 0.10, and 0.04 is determined to be less than 0.10. A retention operation is performed based on this result, retaining the aforementioned five initial feature vector components as valid data components. If the actual consistency ratio obtained in any calculation is greater than 0.10, a filtering operation is performed to completely remove the associated data component. The five retained valid data components are arranged in order to generate a feature weight coefficient vector. The calculated value of 0.04 indicates that there is no logical contradiction in the current pairwise comparison sequence, which satisfies the condition for direct use as the basis for weight allocation.

[0028] S202: Based on the feature weight coefficient vector, call the synchronous dataset of cardiac electrocardiogram signal, body surface temperature and voltage, chamber temperature and voltage, humidity and oxygen concentration data, multiply the values ​​of multiple data channels with the corresponding weight components and accumulate them one by one to obtain the multi-source weighted fusion result; The five weight components contained in the feature weight coefficient vector generated in the preceding steps are extracted, corresponding to the real-time values ​​of the electrocardiogram (ECG), body surface temperature and voltage, chamber temperature and voltage, humidity percentage, and oxygen concentration percentage at the same time node in the synchronized dataset. Substituting the aforementioned parameters, the weight component for the ECG dimension is extracted as 0.45, the weight component for the body surface temperature and voltage dimension as 0.25, the weight component for the chamber temperature and voltage dimension as 0.15, the weight component for the oxygen concentration percentage dimension as 0.10, and the weight component for the humidity percentage dimension as 0.05. The specific values ​​for each item at the current synchronization time node are obtained: the ECG value is 120 mV, the body surface temperature and voltage value is 35 volts, the chamber temperature and voltage value is 25 volts, the oxygen concentration percentage value is 21, and the humidity percentage value is 40. The real-time values ​​extracted from each data channel are multiplied and mapped with their corresponding weight components to generate independent product terms. The heart rate signal value 120 is multiplied by a weighted component 0.45 to obtain an independent product term 54. The body surface temperature voltage value 35 is multiplied by a weighted component 0.25 to obtain an independent product term 8.75. The chamber temperature voltage value 25 is multiplied by a weighted component 0.15 to obtain an independent product term 3.75. The oxygen concentration percentage value 21 is multiplied by a weighted component 0.10 to obtain an independent product term 2.1. The humidity percentage value 40 is multiplied by a weighted component 0.05 to obtain an independent product term 2.0. These five independent product terms generated through multiplication are then summed one by one. The independent product term 54 is then summed with the independent product terms 8.75, 3.75, 2.1, and 2.0, resulting in a multi-source weighted fusion result of 70.6. The weighted fusion result 70.6 comprehensively quantifies the overall physiological load status of the target object at the current time point and serves as the direct data base for subsequent status rating.

[0029] S203: Based on the multi-source weighted fusion result, call the data channel identifier corresponding to the multi-weighted item and perform linear normalization mapping. Match the normalization result with the preset evaluation interval boundary value and perform segment calibration. Combine and encode the calibration segment index with the fusion value to generate physiological evaluation results. The multi-source weighted fusion result 70.6, obtained from the aforementioned steps, is used, and the corresponding multi-weighted data channel identifier code combination sequence is extracted and generated. A linear normalization mapping operation is performed on this multi-source weighted fusion result. The historical maximum and minimum fusion extreme values ​​recorded over 30 consecutive days in the database are retrieved, with the historical maximum fusion extreme value being 100 and the historical minimum fusion extreme value being 0. The difference between the current multi-source weighted fusion result 70.6 and the historical minimum fusion extreme value 0 is calculated to obtain a mapping difference of 70.6. The difference between the historical maximum fusion extreme value 100 and the historical minimum fusion extreme value 0 is calculated to obtain an extreme value interval span of 100. The mapping difference 70.6 and the extreme value interval span of 100 are divided to obtain a linear normalization result of 0.706. A preset evaluation interval boundary value set is obtained. This boundary value set is generated by extracting the physiological fusion data distribution intervals of 500 healthy samples and extracting the corresponding decimals. The first evaluation interval boundary value is set to 0.4, and the second evaluation interval boundary value is set to 0.8. The linear normalization result 0.706 is compared with the boundary values ​​of the preset evaluation interval and matched using interval mapping. It is determined that 0.706 is greater than the first evaluation interval boundary value of 0.4 and less than the second evaluation interval boundary value of 0.8. Based on the numerical comparison result, a segment calibration operation is performed on the linear normalization result, extracting a specific index number corresponding to the middle data range, and labeling the result as a medium physiological state level with a segment index number of 2. The segment index number 2 is extracted. A string concatenation encoding operation is performed between the calibrated segment index number 2 and the original multi-source weighted fusion result 70.6. The segment index number 2 is appended to the header of the identifier code sequence, and then the fusion value 70.6 is appended, finally generating a physiological evaluation result encoding sequence of 270.6. This encoded value represents that the target object is currently in a comprehensive physiological state where all indicators are stable and no abnormal mutations are observed.

[0030] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the physiological evaluation results, obtain the corresponding time series and perform frame extraction analysis on the video images inside the pet transport box. Extract continuous frame sequences and align the timestamps of multiple frames. Match, verify and interpolate the frame timestamps with the time series of physiological evaluation results to obtain a time-synchronized frame sequence index set. Extract the generated physiological evaluation result encoding sequence 270.6, and extract the corresponding baseline physiological time series. Acquire real-time video stream data continuously collected by a wide-angle monitoring camera installed inside the pet transport carrier, setting a fixed sampling frequency of 30 frames per second. Perform continuous frame-by-frame extraction on this real-time video stream data to obtain a total of 150 consecutive frame images lasting 5 seconds, and extract the original generation timestamp of each frame to form a frame time series. Perform a matching and verification operation between the extracted frame time series and the baseline physiological time series, iterating through the original generation timestamp of each frame. Extract the original generation timestamp of a specific frame as 16005 milliseconds, extract the nearest neighbor time node in the baseline physiological time series as 16000 milliseconds, and perform a difference operation between the original generation timestamp 16005 milliseconds and the nearest neighbor time node 16000 milliseconds to obtain an absolute time deviation of 5 milliseconds. If the absolute value of the time deviation is less than the preset tolerance threshold of 5 milliseconds (10 milliseconds), the timestamp of the frame image is forcibly modified to 16000 milliseconds and classified under the corresponding time node. The next target time node in the baseline physiological time series is extracted as 16030 milliseconds, and the original generation timestamp of the frame image closest to this node in the current video stream is obtained as 16045 milliseconds. If the absolute value of the time deviation of this frame is greater than the preset tolerance threshold of 15 milliseconds (10 milliseconds), it is determined that there is a missing valid frame at this node, and interpolation correction is performed at this position. The valid frame data corresponding to 16020 milliseconds and 16040 milliseconds are extracted, and the pixel values ​​at corresponding positions in the two frames are arithmetically averaged to generate a virtual compensation frame. This virtual compensation frame is then used to fill in the baseline time node at 16030 milliseconds. After the above traversal matching and interpolation filling operations, isolated redundant frames are removed, and all frame image data strictly aligned with the baseline physiological time series are retained. The consecutively sorted row number set corresponding to these retained frames is extracted to generate a time-synchronized frame sequence index set.

[0031] S302: Based on the time synchronization frame sequence index set, call the corresponding frame image data, traverse the pixel data of each frame, extract the brightness of multiple pixels, and aggregate the brightness of all pixels under the same timestamp by spatial coordinate mapping, while performing sequential encoding to generate a pixel brightness sequence. The generated time-synchronized frame sequence index set is invoked, and all ordered index row numbers recorded within the index set are extracted to retrieve the corresponding actual frame image data matrix stored in the video buffer database. A full-frame pixel data traversal operation is performed on each retrieved frame image. The image spatial resolution is set to a width of 200 pixels and a height of 150 pixels, obtaining a total of 30,000 independent pixels within a single frame image. The color component values ​​of each independent pixel in the red, green, and blue color channels are extracted. The red component is multiplied by a fixed weight of 0.299, the green component by a fixed weight of 0.587, and the blue component by a fixed weight of 0.114. These three multiplication results are summed to derive the corresponding pixel's grayscale value. The center pixel located at a spatial coordinate of 100 (horizontal coordinate) and 75 (vertical coordinate) is extracted, obtaining its red component as 120, green component as 150, and blue component as 200. These are then multiplied and summed using the aforementioned fixed weights to calculate the grayscale value of the center pixel as 147. All 30,000 grayscale values ​​extracted from a single frame at the same timestamp are aggregated using spatial coordinate mapping. Following an increasing order of horizontal and vertical coordinates in physical space, all scattered brightness values ​​are combined into a two-dimensional brightness data matrix. This two-dimensional brightness data matrix is ​​then subjected to raster scanning-style sequential encoding, extracting brightness values ​​row by row and concatenating them into a one-dimensional linear sequence containing 30,000 numerical nodes. Table 2 shows the sequence distribution of grayscale brightness extraction results for some pixels at the same timestamp.

[0032] Table 2. Example of local pixel brightness mapping

[0033] Extract all the grayscale brightness values ​​that have been quantized as shown in Table 2 and enqueue them along with their spatial coordinates to generate a pixel brightness sequence in a single frame dimension.

[0034] S303: Perform brightness value difference operation on the corresponding coordinate pixels of adjacent frames for the pixel brightness sequence, encode the multiple pixels in the difference result in time order and perform vectorization mapping to generate behavior feature vector; Extract the pixel brightness sequences at each consecutive time point generated. Extract the pixel brightness sequences of the current frame and the previous frame corresponding to two adjacent time points. Perform a difference operation on the brightness values ​​of corresponding coordinate pixels in the two frame sequences. Extract the current grayscale brightness value at the position corresponding to the horizontal coordinate 100 and the vertical coordinate 75 in the current frame sequence as 146, and extract the previous grayscale brightness value at the same coordinate position in the previous frame sequence as 141. Perform a difference operation on the current grayscale brightness value 146 and the previous grayscale brightness value 141 to obtain an absolute brightness change difference of 5. Perform the above difference operation on all 30,000 pixel nodes in a single frame one by one to obtain a difference result array containing 30,000 brightness change differences. Extract this difference result array and determine the numerical relationship between each brightness change difference and the preset silent noise threshold. A preset silent noise threshold of 3 is set. The difference value of 5 is compared to the threshold of 3. If 5 is greater than 3, the corresponding pixel is considered to have undergone real motion change, and the difference value of 5 is retained. If the brightness change difference of a pixel is 2, and 2 is less than 3, the change difference at that location is forcibly reset to zero. All retained pixel differences in the threshold-filtered difference results are sequentially combined and encoded according to the original spatial mapping time order to generate a differential feature code set representing the motion intensity of the entire image within that time interval. This differential feature code set is then vectorized, with the action feature dimensionality reduction mapping dimension set to 256. The differential feature code set containing 30,000 nodes is divided into 256 equal data sub-blocks. For all valid differences within each data sub-block, an arithmetic mean is calculated, and the 256 arithmetic means are extracted as feature dimension components. These 256 feature dimension components are then concatenated sequentially to generate a behavior feature vector with a fixed length of 256 dimensions. This behavioral feature vector directly quantifies the intensity of limb movements and the range of local activity of the target object within a very short time slice.

[0035] Please see Figure 5 The specific steps of S4 are as follows: S401: Align the physiological evaluation results with the behavioral feature vectors according to the corresponding time points, perform index matching based on the timestamp sequence, perform column-wise concatenation operation on the matched numerical vectors according to the field order, and perform interval normalization on the concatenated multi-column values ​​to obtain fused feature data. Extract the generated physiological evaluation results and behavioral feature vectors with timestamps. Obtain the 16000-millisecond timestamp from the preceding baseline physiological time series, extract the physiological evaluation result encoding sequence 2706 at the corresponding time node, and parse the multi-source weighted fusion result 70.6 as a single-dimensional physiological feature component. Simultaneously, retrieve the behavioral feature vector at the same 16000-millisecond timestamp, and extract the 256 action feature dimensionality-reduced components contained within this vector. Perform an index matching operation based on the unified timestamp sequence to confirm that the physiological feature components and the action feature dimensionality-reduced components belong to the same time slice. Perform column-wise concatenation on the matched and verified values, setting the first column as the physiological feature component and the second to 257th columns as the action feature dimensionality-reduced components, combining to generate a multi-column numerical vector containing 257 dimensions. Perform interval normalization on this multi-column numerical vector, and retrieve the extreme value benchmarks for each dimension from historical data. Extract the historical maximum benchmark value of the physiological feature dimension as 100 and the minimum benchmark value as 0. Extract the historical maximum benchmark value of the first action feature dimension as 255 and the minimum benchmark value as 0. The difference between the current fusion result 70.6 and 0 is 70.6, and the difference between 100 and 0 is 100. Dividing 70.6 by 100 yields a normalized physiological value of 0.706. The first action feature value is extracted as 120, and the difference between this value and 0 is 120. Dividing this by the extreme value difference between 255 and 0 (255) yields a normalized action value of 0.470. This process of subtracting the minimum value and dividing by the extreme value difference is performed on all 257 dimensions, uniformly compressing all dimension values ​​to a distribution range of 0 to 1. The resulting 257 floating-point values ​​are then arranged in their original order to generate fused feature data. The resulting 0.706 and 0.470 represent the relative size distribution of each heterogeneous feature at the global scale, eliminating the interference of dimensional differences on the weight calculation of the classifier.

[0036] S402: Calculate the classification margin metric for multiple sample vectors based on fused feature data, group and map the sample vectors according to the category labels, calculate the candidate set of support vectors for multiple groups of sample vectors, and select the vector set that meets the margin constraint to establish the support vector set; A training sample set containing 1000 sets of historical fused feature data is retrieved, and the pre-labeled category label attached to each set of samples is extracted. A support vector machine classification model is constructed, and the radial basis function is called to perform nonlinear spatial mapping calculation on the previously generated 257-dimensional fused feature data, mapping it to a high-dimensional feature space to enhance the linear separability of the features. A grouping mapping operation is performed based on the normal state category label and the abnormal state category label attached to the sample set, dividing the training sample set into normal feature clusters and abnormal feature clusters. A classification margin metric is calculated for the multi-sample vectors contained in the two feature clusters, measuring the vertical Euclidean distance of each multi-sample vector to the preset initial classification boundary. The distance measurement value from a specific normal multi-sample vector to the initial classification boundary is set to 1.5, and the distance measurement value from a specific abnormal multi-sample vector to the initial classification boundary is set to 1.2. A preset margin constraint threshold is obtained, which is calculated by extracting the average value of the support vector distance distribution in the historical best classification model and set to 2.0. The extracted distance measurement value of 1.5 is compared with the threshold value of the interval constraint 2.0, and it is determined that 1.5 is less than 2.0. The extracted distance measurement value of 1.2 is then compared with the threshold value of the interval constraint 2.0, and it is determined that 1.2 is less than 2.0. For sample vectors whose distances are less than the constraint threshold, an extraction and filtering operation is performed, and they are added to the candidate support vector set. Redundant sample points with distances greater than 2.0 are removed, and the final set of sample points that meet the conditions is established as the core support vector set. The distance values ​​of 1.5 and 1.2 that satisfy the constraint conditions represent sample points located in the sensitive edge zone of the classification boundary, constituting key support anchor points that determine the final orientation of the classification hyperplane.

[0037] S403: Perform spatial partitioning operations on the multi-sample vectors in the fused feature data based on the support vector set, construct hyperplane parameters, perform symbol mapping calculation on the multi-sample vectors, match the mapping results with the category codes, and generate state evaluation results; Extract all multi-sample vector data contained within the established support vector set. Perform spatial partitioning operations based on these multi-sample vectors located in the boundary zone. Set the hyperplane parameters to include weight normal vectors and translation intercept scalars, and use a gradient descent optimization mechanism to perform iterative update operations on the hyperplane parameters. In the iterative calculation, set the loss function to be the arithmetic sum of the inverse of the margin width and the misclassified sample penalty term. Adjust the weight values ​​by calculating the direction of the partial derivative of the loss function with respect to the weight normal vector to maximize the distance between each support vector and the partition boundary. After 500 iterations, extract the final fixed hyperplane parameters. Substitute the fused feature data containing 257 dimensions generated at the current time node in the previous steps, perform a dot product operation with the fixed weight normal vectors, and accumulate the results to obtain an inner product of 3.5. Extract the determined translation intercept scalar as -1.2, and sum the inner product result 3.5 with the translation intercept scalar -1.2 to calculate the spatial projection position value of 2.3. For the spatial projection position value 2.3, a sign mapping calculation is performed, determining that 2.3 is greater than 0. According to the positive / negative determination rule, values ​​greater than 0 are uniformly mapped to a positive sign code 1. If the obtained spatial projection position value is less than 0, it is mapped to a negative sign code -1. This positive sign code 1 is extracted and imported into a preset pet physiological and behavioral comprehensive assessment dictionary for corresponding matching operations. The extracted positive sign code 1 corresponds to the category code identifier representing a healthy, active, and stable period. This category code identifier is used as the final output to generate the state assessment result for the current time slice. The aforementioned spatial projection position value 2.3 and mapping code 1 indicate that the target pet's current multidimensional fusion features strictly fall within the normal state high-dimensional feature space range, excluding any tendency to judge physiological or behavioral abnormalities.

[0038] Please see Figure 6 The specific steps of S5 are as follows: S501: Compare the status assessment results with the preset hierarchical rule configuration table, call multiple status indicators in the status assessment results and compare them with the corresponding level ranges in the hierarchical rule configuration table, match and mark the rule index identifiers according to the ranges where the indicators fall, and generate a status hierarchical index sequence. Extract the generated status assessment results, including the comprehensive status category code identifier 1, and the corresponding physiological evaluation fusion value of 70.6 and normalized action value of 0.470, among other status indicators. Retrieve a pre-built preset grading rule configuration table, which is set according to the distribution patterns of historical live pet transportation status data. Extract the boundary thresholds of the physiological evaluation level intervals from the configuration table, setting the first interval to 0-60, the second interval to 60-80, and the third interval to 80-100. Compare the physiological evaluation fusion value of 70.6 in the status assessment results with each level interval. If 70.6 is greater than 60 and less than 80, the indicator is determined to fall into the second interval. Extract the behavior level intervals from the configuration table, setting the low-frequency interval to 0-0.3, the mid-frequency interval to 0.3-0.6, and the high-frequency interval to 0.6-1.0. Compare the normalized action value of 0.470 with the aforementioned action intervals, determining that 0.470 is greater than 0.3 and less than 0.6, thus the indicator falls into the mid-frequency interval. Based on the specific distribution of the above indicators falling within the interval range, an addressing association operation is performed to extract the rule index identifiers that commonly correspond to the second physiological interval and the mid-frequency behavioral interval. The matching rule index identifier value is determined to be 2. This identifier value of 2 is arranged in chronological order to generate a state-level index sequence. Table 3 shows the interval division of some indicators.

[0039] Table 3 Preset Hierarchical Rule Configuration Table

[0040] Table 3 shows the distribution of the label values ​​corresponding to the physiological and behavioral index intervals. The obtained rule index label value of 2 indicates that the target pet is in a physiologically stable state without excessive behavior, which serves as the basis for subsequent hardware parameter adjustments.

[0041] S502: Based on the state hierarchical index sequence, call the associated fields in the hierarchical rule configuration table, extract the ventilation speed corresponding to each index identifier and perform normalization conversion, map the speed corresponding to multiple indices to a unified scale sequence, and at the same time associate the encoded value of the interactive prompt field to generate a ventilation speed parameter encoding set; Receive the generated state hierarchy index sequence and extract the rule index identifier value 2 for the current time node. Based on this value, call the hardware control association field in the hierarchy rule configuration table. Extract the theoretical ventilation speed parameter value corresponding to index identifier 2 as 1500 revolutions per minute. Obtain the physical limit maximum speed upper limit benchmark value of the ventilation fan hardware as 3000 revolutions per minute, and extract the minimum speed lower limit benchmark value as 0 revolutions per minute. Calculate the ratio between the theoretical ventilation speed parameter value 1500 and the limit maximum speed upper limit benchmark value 3000, dividing 1500 by 3000 to obtain a proportionality coefficient of 0.50. Multiply this proportionality coefficient 0.50 by the maximum digital control duty cycle value of 255 to calculate the decimal quantized value of the ventilation speed parameter as 127. Perform the same conversion operation on the indices under multiple consecutive time nodes, mapping multiple sets of speed values ​​to a unified scaling sequence. Extract the non-numerical text data of the interactive prompt synchronized with index identifier 2 in the hierarchy rule configuration table, and read this text data as stable. A quantization process was developed for this non-numerical text. The pre-defined letter code N corresponding to the text was extracted, and its decimal encoding in the international standard code library was found to be 78. This decimal encoding value 78 was then used as the encoding value for the interactive prompt field to perform an association mapping operation. A correspondence was established between the aforementioned ventilation speed parameter's decimal quantization value 127 and the interactive prompt field's encoding value 78, generating a ventilation speed parameter encoding set. The quantization value 127 and the encoding value 78 represent the transformation of the abstract state into execution parameters directly read by the underlying hardware.

[0042] S503: Based on the ventilation speed parameter encoding set, the speed parameter sequence and interactive prompt encoding are concatenated, the check bits are calculated and the frame header identifier is filled according to the data frame structure order, and the concatenated data sequence is integrated with binary encoding to generate travel management results; The constructed ventilation speed parameter encoding set is retrieved, and the speed parameter sequence value of 127 is extracted from this encoding set, along with the corresponding interactive prompt field encoding value of 78. The standard data frame structure order rules specified by the underlying communication protocol are obtained, and the memory arrangement order required by the rules is extracted as frame header, control parameters, prompt encoding, and check bit. According to this data frame structure order, a field concatenation operation is performed on the speed parameter sequence and the interactive prompt encoding, placing the speed parameter value 127 at the beginning of the data field, followed by the prompt field encoding value 78 in a continuous memory address arrangement. A checksum calculation operation is performed on the concatenated data field. An arithmetic sum operation is performed on the speed parameter value 127 and the prompt field encoding value 78, resulting in a sum of 205. The preset redundancy modulo base is extracted as 256, and the sum 205 is divided by this base 256, yielding a remainder of 205. This remainder 205 is set as the dedicated check bit for the current communication frame. The globally preset frame header identifier is fixed at 170, which is hardcoded by the hardware communication interface protocol. The frame header identifier 170 is filled into the starting address position of the splicing sequence, and the calculated check bit 205 is appended to the end of the sequence, generating a decimal array containing 170, 127, 78, and 205. A binary encoding integration operation is performed on this spliced ​​data sequence. The frame header identifier decimal number 170 is converted to an eight-bit binary sequence 10101010, the rotation speed parameter 127 is converted to a binary sequence 01111111, the prompt code 78 is converted to 01001110, and the check bit 205 is converted to 11001101. These four sets of eight-bit binary sequences are seamlessly concatenated in sequence to generate a binary level pulse data stream with a total length of thirty-two bits, which serves as the final travel management result sent to the pet transport box's main control chip. This data stream represents a set of machine instructions for performing a closed-loop intervention on the target cabin's microenvironment.

[0043] Please see Figure 7 A pet travel end-to-end management system based on multi-source perception and real-time interaction includes: The data analysis module collects pet heart rate signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, combines preset time span thresholds to remove out-of-bounds time differences, generates a synchronous dataset and transmits it to the physiological assessment module. The physiological assessment module assigns weights to the synchronous dataset, multiplies and sums the heart rate electrical signal, body surface temperature and voltage, chamber temperature and voltage, humidity, and oxygen concentration with their corresponding weight coefficients, generates physiological assessment results, and transmits them to the behavior extraction module. The behavior extraction module acquires video images inside the transport box at the time corresponding to the physiological evaluation results and performs adaptive frame extraction in combination with the physiological evaluation scores. It subtracts the brightness of pixels at corresponding coordinate positions in adjacent frames and extracts pixels with brightness changes, generates a behavior feature vector, and passes it to the state classification module. The state classification module concatenates and aggregates the physiological evaluation results with the behavioral feature vectors and inputs them into the support vector machine for classification. It performs mapping and classification calculations on the feature fusion results based on the spatial partitioning hyperplane, generates state evaluation results, and transmits them to the travel management module. The travel management module compares the status assessment results with the preset hierarchical rule configuration table, extracts the environmental ventilation speed parameters and interactive prompt codes that match the status assessment results based on the association mapping logic, and encapsulates them to generate travel management results.

[0044] The above are merely specific embodiments 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 managing the entire pet travel process based on multi-source sensing and real-time interaction, characterized in that, Includes the following steps: S1: Collect pet heartbeat electrical signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, combine them with preset time span thresholds to remove out-of-bounds time differences, and generate a synchronous dataset; S2: Weight the synchronous dataset by multiplying the electrocardiogram signal, body surface temperature and voltage, chamber temperature and voltage, humidity and oxygen concentration by their respective weight coefficients and summing them to generate physiological evaluation results. S3: Obtain the video image inside the transport box at the time corresponding to the physiological evaluation result and perform adaptive frame extraction in combination with the physiological evaluation score. Subtract the brightness of the pixels at the corresponding coordinate positions of adjacent frames and extract the pixels with brightness changes to generate a behavior feature vector. S4: The physiological evaluation results are concatenated and aggregated with the behavioral feature vectors and input into a support vector machine for classification. The feature fusion results are mapped and classified based on the spatial partitioning hyperplane to generate state evaluation results. S5: Compare the status assessment results with the preset hierarchical rule configuration table, extract the environmental ventilation speed parameters and interactive prompt codes that match the status assessment results based on the association mapping logic, and encapsulate them to generate travel management results.

2. The method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 1, characterized in that, The synchronized dataset includes cardiac electrocardiogram sequences, body surface temperature and voltage sequences, oxygen concentration change sequences, and humidity change sequences. The physiological evaluation results include physiological scores, physiological stability indices, and multi-parameter fusion indices. The behavioral feature vectors include brightness change distributions, activity activity features, and pixel difference statistics. The state assessment results include state category labels, state confidence scores, and classification boundary mapping parameters. The travel management results include ventilation speed adjustment parameters, environmental control instruction sets, and interactive prompt codes.

3. The method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collects pet heartbeat electrical signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, timestamps the output values ​​of multiple sensor channels in chronological order, and aligns the values ​​at the same time index position to obtain multi-source time-stamped data; S102: Based on the multi-source time-labeled data, call the multi-channel timestamp sequence, calculate the difference between adjacent timestamps, compare the obtained difference with the preset time span threshold item by item, index and mark the positions where the difference exceeds the time span threshold and remove the corresponding data rows to obtain the time difference constraint filtering dataset; S103: Based on the time difference constraint, filter the dataset, call the remaining data row index set, perform unified time series rearrangement on the multi-channel values, and perform index compression on the missing positions to obtain the synchronized dataset.

4. The method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 3, characterized in that, The time span threshold is determined by analyzing the hardware reference sampling period parameters of the multi-source sensor and the fixed delay time constant of the communication link. The first and last timestamp span values ​​of the initial sample sequence are extracted and divided by the total number of samples to calculate the mean compensation amount. The mean compensation amount is then combined with the sum of the reference sampling period parameters and the fixed delay time constant to determine the threshold.

5. The method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the multi-data input hierarchical analysis method of the synchronous dataset, weight allocation is performed, a pairwise comparison sequence is constructed for multiple data dimensions and the feature vector is solved, the multiple components of the feature vector are normalized and data components that do not meet the preset consistency ratio threshold are filtered out, and feature weight coefficient vector is generated. S202: Based on the feature weight coefficient vector, call the synchronous dataset of heartbeat signal, body surface temperature and voltage, chamber temperature and voltage, humidity and oxygen concentration data, multiply the values ​​of multiple data channels with the corresponding weight components and accumulate them one by one to obtain the multi-source weighted fusion result. S203: Based on the multi-source weighted fusion result, call the data channel identifier corresponding to the multi-weighted item and perform linear normalization mapping, match the normalization result with the preset evaluation interval boundary value and perform segment calibration, combine the calibration segment index with the fusion value and encode to generate physiological evaluation result.

6. The method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 5, characterized in that, The consistency ratio threshold is determined by calling the matched average random consistency index through the order constant of the pairwise comparison sequence, and performing multiplicative scaling calculation in combination with the judgment fault tolerance constant. The numerical limit obtained by scaling calculation is then summed and compensated with the underlying data scaling factor to determine the final value.

7. The method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the physiological evaluation results, obtain the corresponding time series and perform frame extraction analysis on the video images inside the pet transport box. Extract continuous frame sequences and align the timestamps of multiple frames. Match, verify, and interpolate the frame timestamps with the time series of the physiological evaluation results to obtain a time-synchronized frame sequence index set. S302: According to the time synchronization frame sequence index set, call the corresponding frame image data, traverse the pixel data of each frame, extract the brightness of multiple pixels, and aggregate the brightness of all pixels under the same timestamp by spatial coordinate mapping, while performing sequential encoding to generate a pixel brightness sequence. S303: Perform brightness value difference operation on the corresponding coordinate pixels of adjacent frames for the pixel brightness sequence, encode the multiple pixels in the difference result in time order and perform vectorization mapping to generate behavioral feature vectors.

8. The method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Align the physiological evaluation results with the behavioral feature vector according to the corresponding time points, perform index matching based on the timestamp sequence, perform column-wise concatenation operation on the matched numerical vector according to the field order, and perform interval normalization on the concatenated multi-column values ​​to obtain fused feature data. S402: Based on the fused feature data, perform classification interval measurement calculation on multiple sample vectors, group and map the sample vectors according to the category labels, calculate the support vector candidate set for multiple groups of sample vectors, and filter the vector set that meets the interval constraint to establish the support vector set. S403: Based on the support vector set, perform spatial partitioning operation on the multi-sample vectors in the fused feature data, construct hyperplane parameters, perform symbol mapping calculation on the multi-sample vectors, match the mapping results with the category codes, and generate state evaluation results.

9. A method for managing the entire pet travel process based on multi-source perception and real-time interaction according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: The state assessment result is compared with the preset hierarchical rule configuration table. Multiple state indicators in the state assessment result are compared with the corresponding level ranges in the hierarchical rule configuration table. The rule index identifier is matched and marked according to the range of the indicator, and a state hierarchical index sequence is generated. S502: Based on the state classification index sequence, call the associated fields in the classification rule configuration table, extract the ventilation speed corresponding to each index identifier and perform normalization conversion, map the speeds corresponding to multiple indices to a unified scale sequence, and at the same time associate the encoding values ​​of the interactive prompt field to generate a ventilation speed parameter encoding set. S503: Based on the ventilation speed parameter encoding set, the speed parameter sequence and interactive prompt encoding are concatenated, the check bit is calculated and the frame header identifier is filled according to the data frame structure order, and the concatenated data sequence is integrated with binary encoding to generate travel management results.

10. A pet travel end-to-end management system based on multi-source sensing and real-time interaction, characterized in that, The system is used to implement the pet travel end-to-end management method based on multi-source perception and real-time interaction as described in any one of claims 1-9, the system comprising: The data analysis module collects pet heart rate signals, body surface temperature and voltage, box temperature and voltage, oxygen concentration and humidity data, combines preset time span thresholds to remove out-of-bounds time differences, generates a synchronous dataset and transmits it to the physiological assessment module. The physiological assessment module assigns weights to the synchronous dataset, multiplies and sums the heart rate electrocardiogram, body surface temperature and voltage, chamber temperature and voltage, humidity, and oxygen concentration with their corresponding weight coefficients, generates physiological assessment results, and transmits them to the behavior extraction module. The behavior extraction module acquires the video image inside the transport box at the time corresponding to the physiological evaluation result and performs adaptive frame extraction in combination with the physiological evaluation score. It subtracts the brightness of the pixels at the corresponding coordinate positions of adjacent frames and extracts the pixels with brightness changes, generates a behavior feature vector, and passes it to the state classification module. The state classification module concatenates and aggregates the physiological evaluation results with the behavioral feature vectors and inputs them into a support vector machine for classification. It performs mapping and classification calculations on the feature fusion results based on the spatial partitioning hyperplane, generates state evaluation results, and transmits them to the travel management module. The travel management module compares the status assessment results with the preset hierarchical rule configuration table, extracts the environmental ventilation speed parameters and interactive prompt codes that match the status assessment results based on the association mapping logic, and encapsulates them to generate travel management results.

Images