Sheep weight intelligent monitoring method and system based on drinking water behavior

Abstract

This invention discloses a method and system for intelligent monitoring of sheep weight based on drinking behavior, comprising the following steps: S1, continuously monitoring weight sensor data and triggering drinking monitoring; S2, performing time-series capture and identifier parsing on the radio frequency response signal to extract valid electronic identifiers to form sheep identity results; S3, performing time-series unfolding on the weight sensor data and implementing segmented segmentation to screen stable carrying areas; S4, performing moving median filtering on the weight sample values ​​and constraining residual fluctuations to obtain stable weight values; S5, performing dynamic time-registration registration and solving for the optimal registration path to generate individual binding results; S6, performing structured packaging and link transmission organization, and sending to a remote server; S7, performing historical aggregation and identifying abnormal offset segments, outputting weight monitoring results and abnormal warning information. This invention can realize intelligent weight monitoring of sheep during the drinking process, improving the stability of weight measurement.

Description

Technical Field

[0001] This invention relates to the field of weight monitoring technology, and in particular to an intelligent method and system for monitoring the weight of sheep based on drinking behavior. Background Technology

[0002] In large-scale sheep farming, weight data is a crucial indicator of growth status, feeding effectiveness, and health changes. Farmers typically rely on weight changes for group management, feeding adjustments, and screening for abnormal conditions. Common existing methods for obtaining sheep weight include manually driving them to weighbridges, platform scales, or portable weighing devices, or setting up dedicated weighing channels within the sheepfold to identify and collect weight data as sheep pass through designated areas.

[0003] With the development of IoT and RFID technologies, some automated weighing solutions have begun to combine weight sensors with RFID devices to acquire individual sheep weight data at feeding areas, passageways, or designated standing areas. While these solutions reduce the burden of manual recording to some extent, their implementation still typically relies on sheep entering a specific weighing area, and the identification, weight acquisition, and data upload during weighing are mostly implemented in series as independent steps, making the overall monitoring process still biased towards discrete measurement.

[0004] Existing technologies have at least the following shortcomings: First, existing weighing methods often require herding, restricting passageways, or setting up dedicated weighing areas, which can easily cause stress in sheep and make it difficult to deeply integrate with their daily natural behavior. Second, the weight response in existing solutions is easily affected by changes in posture such as head raising, head retraction, body shifting, and limb movement, resulting in significant weight fluctuations and making it difficult to stably extract the effective load-bearing segment from continuous weight responses. Third, existing identification processes and weight measurement processes lack a close temporal correlation mechanism, which can easily lead to inaccurate correspondence between electronic tag responses and actual weight results. Fourth, existing systems lack sufficient structured organization and remote transmission of single weighing results to subsequent monitoring data, making it difficult to form a continuous and stable weight monitoring link.

[0005] Therefore, how to provide a method and system for intelligent monitoring of sheep weight based on drinking behavior is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent method and system for monitoring sheep weight based on drinking behavior. This invention fully utilizes technologies such as weight sensing, radio frequency identification, CUSUM change point detection, sliding median filtering, least squares fitting, and dynamic time warping registration. It describes in detail the implementation methods for monitoring triggering, identification, stable carrying range screening, stable weight value calculation, individual binding, and continuous weight evolution analysis during the natural drinking process of sheep. It has the advantages of low stress interference, high monitoring frequency, strong stability of weight measurement, high accuracy of individual binding, and strong continuous early warning capability.

[0007] A method for intelligent monitoring of sheep weight based on drinking behavior according to an embodiment of the present invention includes the following steps: S1. Continuously monitor the weight sensor data of the drinking water terminal and trigger the drinking water monitoring state according to the preset threshold; S2. Under drinking water monitoring status, the radio frequency response signal in the drinking trough identification area is captured in time and the identifier is parsed to extract the valid electronic identifier of the current drinking process and form the sheep's identity result. S3. Perform time-series unfolding on the weight sensing data, and use the variable point detection method to segment the continuous interval of weight fluctuation, eliminate the unsteady load-bearing section caused by attitude change, and screen out the stable load-bearing section. S4. Perform sliding median filtering on the weight sampling values ​​within the stable bearing section, and use least squares fitting to construct the bearing trend line. Constrain the residual fluctuations that deviate from the bearing trend line to obtain the stable weight value. S5. Based on stable weight values, perform dynamic time warping and registration on the corresponding time intervals, and solve for the optimal registration path under the constraints of a preset matching window to generate individual binding results; S6. Perform structured packaging and link transmission organization on the individual binding results, and send the corresponding monitoring data to the remote server through the communication module; S7. Perform historical aggregation on the received monitoring data on the remote server, and use the moving average method to construct the individual weight evolution curve, identify abnormal offset segments, and output weight monitoring results and abnormal warning information.

[0008] Optionally, S1 specifically includes: S11. Perform continuous sampling on the weight sensing data output by the drinking water terminal to form a weight response sequence arranged in time order; S12. Perform continuous interval scanning on the weight response sequence, extract the static dwelling interval where the weight fluctuation amplitude is lower than the preset fluctuation threshold and the duration reaches the preset length, and solve the mean value of the weight sampled values ​​in the static dwelling interval to obtain the static bearing baseline as a reference for the weight of the drinking water terminal in the static state. S13. Perform difference expansion between each weight sample value in the weight response sequence and the static load baseline to obtain the weight offset at each time point, and construct a static threshold response based on the comparison result between the weight offset and the preset trigger increment threshold. S14. Perform continuous dwell confirmation on the static threshold response. When the weight offset continuously exceeds the preset static threshold and the continuous over-limit time reaches the preset duration threshold, the drinking water monitoring state is triggered. The preset static threshold is the weight trigger threshold formed by superimposing the preset trigger increment on the static bearing baseline. The drinking water monitoring state is the activation state in which the drinking water terminal switches from weight monitoring to identity recognition and subsequent weight measurement.

[0009] Optionally, S2 specifically includes: S21. Under drinking water monitoring status, control the radio frequency identification device in the drinking trough identification area to perform periodic gating excitation according to the preset polling rhythm, so that the electronic tag entering the identification area emits the corresponding radio frequency response in each polling cycle; S22. Perform time slot occupancy analysis on each radio frequency response along the polling cycle to determine the response time slot position of each electronic identifier in the current polling cycle, forming a candidate response set arranged in time slot order; S23. Use a time slot anti-collision method to resolve conflicts in the candidate response set, re-polling and allocating overlapping response time slots with concurrent occupancy, and retaining independent response time slots without concurrent occupancy as candidate identifier time slots. S24. Perform integrity checks on the response results within the candidate identifier time slots, remove abnormal response time slots corresponding to incomplete responses, interrupted responses, and abnormal occupancy, and retain valid response time slots that meet the continuous response conditions. S25. Perform continuous dominance locking on the electronic identifiers corresponding to valid response time slots, and determine the electronic identifiers that stably occupy positions within multiple consecutive polling cycles as the target electronic identifiers for the current drinking process, forming sheep identification results, specifically including: Extract the electronic identifiers corresponding to each valid response time slot along the continuous polling cycle, and construct the electronic identifier response sequence according to the polling order; The number of times each electronic identifier appears in the electronic identifier response sequence within the continuous polling period is accumulated, and the continuous occupancy length of the corresponding effective response time slot of each electronic identifier is simultaneously counted. Electronic identifiers that appear a number of times reaching a preset dominant frequency threshold and whose continuous occupancy length reaches a preset occupancy length threshold are identified as dominant candidate identifiers. When there are multiple dominant candidate identifiers, a stability comparison is performed on the continuous occupancy segments corresponding to each dominant candidate identifier, and the dominant candidate identifier with the smallest fluctuation in the continuous occupancy segment is retained as the target electronic identifier. The target electronic tag is identified as the identification tag corresponding to the current drinking process, thus forming the sheep identification result.

[0010] Optionally, S3 specifically includes: S31. The weight sensing data is unfolded according to the sampling time to form a weight fluctuation sequence, and the difference sequence between adjacent sampled values ​​is calculated along the weight fluctuation sequence. S32. The CUSUM change point detection method is used to perform cumulative offset search on the differential sequence. When the cumulative offset result exceeds the preset change threshold, the load change point is locked. The weight fluctuation sequence is divided into multiple continuous load segments according to each load change point. S33. Perform local slope fitting and segment range calculation on each continuous load-bearing section. Continuous load-bearing sections with local slope exceeding a preset slope threshold or segment range exceeding a preset fluctuation threshold are identified as unsteady load-bearing sections and are removed. The local slope fitting is to perform least squares fitting on the weight sampled values ​​in the continuous load-bearing section along the sampling time axis and extract the fitting slope. The segment range calculation is to calculate the difference between the maximum weight sampled value and the minimum weight sampled value in the continuous load-bearing section. S34. Perform a continuous length check on the remaining continuous load-bearing sections after stripping the unsteady load-bearing sections, and retain the continuous load-bearing sections whose continuous length reaches the preset length threshold as stable load-bearing sections.

[0011] Optionally, S32 specifically includes: S321. Perform positive and negative accumulation on each difference value in the difference sequence according to the sampling order to form a bidirectional cumulative offset sequence corresponding to each sampling time. S322. Perform cumulative offset search along the bidirectional cumulative offset sequence, and determine the sampling position where the absolute value of the cumulative offset continues to increase and exceeds the preset mutation threshold as the candidate mutation position. The preset mutation threshold is the cumulative offset judgment threshold that distinguishes between stable load changes and attitude mutation disturbances. S323. Using each candidate mutation location as the center, search forward along the sampling time axis to find the starting position of the cumulative offset, and search backward along the sampling time axis to find the falling position of the cumulative offset, and lock the sampling location where the absolute value of the cumulative offset reaches the extreme value as the carrying mutation point. S324. According to the sequential distribution of each load change point in the weight fluctuation sequence, the weight fluctuation sequence is divided into intervals, and the weight response interval between adjacent load change points is divided into continuous load segments.

[0012] Optionally, S4 specifically includes: S41. Along the time axis of the stable bearing section, the weight sample values ​​are normalized by sliding window midpoint. In each sliding window, the weight sample values ​​that deviate from the window midpoint and exceed the preset local fluctuation threshold are reset to the corresponding window midpoint, forming a normalized weight sequence with suppressed local disturbance. S42. Using the sampling time of the regular weight sequence as the independent variable and the weight value as the dependent variable, the least squares fitting is used to perform overall trend approximation on the stable bearing section, and the slope parameter and intercept parameter of the bearing trend line are solved to make the weight response in the regular weight sequence concentrate towards the bearing trend line. S43. Apply residual offset constraints to the regular weight sequence along the bearing trend line, establish offset tolerance boundaries on both sides of the bearing trend line, and apply convergence restrictions to weight sample values ​​exceeding the offset tolerance boundaries to form a constrained weight sequence, specifically including: A two-sided tolerance expansion is performed on the bearing trend line along the time axis of the stable bearing section. Constraint boundaries corresponding to the positive offset tolerance and the negative offset tolerance are constructed on both sides of the bearing trend line to form a two-sided tolerance constraint band surrounding the bearing trend line. Trend alignment is performed on the regular weight sequence along the double-sided tolerance constraint band, and the deviation of each weight sample value relative to the bearing trend line is projected onto the tolerance coordinates of the corresponding sampling time to form a residual offset sequence arranged in time sequence. Perform a two-sided over-limit scan on the residual offset sequence to separate the upper offset sampling points that cross the positive tolerance boundary and the lower offset sampling points that cross the reverse tolerance boundary, and mark each over-limit sampling point as exceeding the limit. The Huber constraint function is used to apply convergence constraints to the weight sample values ​​corresponding to the over-limit marker, compressing the upper offset sample points into the positive tolerance boundary and pulling the lower offset sample points into the reverse tolerance boundary, thus forming a boundary-constrained weight response. Sequence reconstruction is performed on the boundary-constrained weight response along the original sampling time sequence, so that the weight sampled values ​​after convergence constraint are re-embedded into the time position corresponding to the regular weight sequence to form a constrained weight sequence. S44. Perform convergence fitting confirmation on the restricted weight sequence. When the slope change and intercept change of two adjacent fitting results both fall within the preset convergence threshold, lock the convergence bearing trend line. S45. Perform stable weight center locking on the restricted weight sequence along the convergent bearing trend line, and determine the center value with the smallest total residual offset and the largest continuous residence length as the stable weight value.

[0013] Optionally, S5 specifically includes: S51. Extract the identification time series corresponding to the sheep's identity results and the measurement time series corresponding to the stable weight values, and apply time axis constraints to the identification time series and measurement time series according to the preset matching window to form the time series to be registered. S52. Construct a dynamic time warping cost matrix for the time series to be registered, and recursively accumulate the registration cost along the row and column directions of the cost matrix to form the temporal registration cost distribution between each identification time node and each measurement time node. S53. Under the constraints of the preset matching window, perform path search on the temporal registration cost distribution, block out-of-window paths that exceed the preset matching window, and find the optimal registration path with the minimum cumulative registration cost among the remaining feasible paths. S54. Map the identification time node and the measurement time node one-to-one along the optimal registration path, and perform association locking on the sheep identity results and stable weight values ​​that fall within the same path mapping unit to generate individual binding results.

[0014] Optionally, S6 specifically includes: S61. The individual binding results are structured and arranged according to the preset data organization order to form monitoring data units corresponding to the current drinking water process; S62. Perform frame encapsulation on the monitoring data unit, and add a frame header identifier and a check field to the outside of the monitoring data unit to form a data frame to be sent. S63. Perform sending queue arrangement on the data frames to be sent, write each data frame to be sent into the sending queue in the order of sending, and assign a corresponding sending sequence number to each data frame to be sent. S64. The drive communication module sends each data frame to be sent sequentially according to the sending sequence number, and performs verification on the returned link response result. When the link response result indicates that the sending has failed, the corresponding data frame to be sent is retransmitted until the data transmission is completed. The communication module is used to send the data frame to be sent to the remote server and receive the link response result. The remote server is a server device set outside the drinking water terminal and establishing a data communication connection with the communication module.

[0015] Optionally, S7 specifically includes: S71. On the remote server, perform historical merging and time normalization on the time-series monitoring units along the monitoring time axis, and rearrange the time-series monitoring units corresponding to the same sheep into a continuous weight monitoring sequence according to the monitoring time. S72. Implement moving average recursion on the continuous weight monitoring sequence, perform mean update and sequence smoothing on the weight monitoring values ​​in each sliding window, and form the individual weight evolution curve corresponding to the monitoring time axis. S73. Perform offset trajectory expansion along the individual weight evolution curve, map the offset between the weight monitoring value at each monitoring time and the corresponding moving average to the monitoring time axis, and implement anomaly locking for monitoring segments with continuous offset directions and offset amplitudes that continuously exceed the preset offset threshold. S74. Perform type merging and result mapping on the monitoring segments after anomaly locking, organize the anomaly monitoring segments and individual weight evolution curves into weight monitoring results, and output the corresponding anomaly warning information.

[0016] An intelligent weight monitoring system for sheep based on drinking behavior, according to an embodiment of the present invention, includes: The monitoring trigger module is used to continuously monitor the weight sensor data of the drinking water terminal and trigger the drinking water monitoring status according to the preset threshold. The identification module is used to capture and analyze the radio frequency response signal in the identification area of ​​the drinking trough in a timely manner during drinking water monitoring, extract the valid electronic identifier of the current drinking process, and form the sheep's identification result. The segment screening module is used to perform time-series unfolding of weight sensing data and to segment the continuous interval of weight fluctuation using a change point detection method, thereby eliminating unsteady load-bearing segments caused by attitude changes and screening out stable load-bearing segments. The weight calculation module is used to perform sliding median filtering on the weight sample values ​​within the stable bearing section, and to construct the bearing trend line using least squares fitting. It also constrains residual fluctuations that deviate from the bearing trend line to obtain stable weight values. The binding and registration module is used to perform dynamic time-warped registration of the corresponding time segment based on stable weight values, and to solve for the optimal registration path under the constraints of a preset matching window to generate individual binding results; The unit construction module is used to perform structured packaging and link transmission organization of individual binding results, and send the corresponding monitoring data to the remote server through the communication module; The analysis and early warning module is used to perform historical aggregation on the received monitoring data on a remote server, construct an individual weight evolution curve using the moving average method, identify abnormal offset segments, and output weight monitoring results and abnormal early warning information.

[0017] The beneficial effects of this invention are: First, this invention embeds the weight monitoring process into the sheep's natural drinking behavior, triggering the monitoring state with the weight sensor data from the drinking terminal, and combining it with radio frequency identification to complete individual identity verification. It no longer relies on manual driving, centralized restraint, or dedicated weighing channels, thus improving the autonomy of weight data acquisition and the frequency of monitoring.

[0018] Secondly, this invention does not simply take values ​​from the original weight readings. Instead, it introduces change point detection, stable load-bearing section screening, sliding median normalization, least squares fitting, and convergent weight center locking around the weight fluctuation sequence. This process removes the unsteady load-bearing sections caused by attitude changes and constrains the residual fluctuations within the stable load-bearing sections, thereby enhancing the reliability of the stable weight value solution process.

[0019] Finally, by binding the identification results with stable weight values ​​to individuals and performing structured packaging and link transmission organization on the monitoring data, this invention enables the weight results generated during a single drinking process to be stably connected to the remote monitoring link, facilitating subsequent continuous weight tracking and abnormal change analysis, thereby improving the data continuity, individual correspondence accuracy, and remote management application capabilities of sheep weight monitoring. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a method for intelligent monitoring of sheep weight based on drinking behavior proposed in this invention; Figure 2 This is a flowchart illustrating the structured packaging and link transmission of monitoring data for a method for intelligent monitoring of sheep weight based on drinking behavior proposed in this invention. Figure 3 This is a module structure diagram of an intelligent sheep weight monitoring system based on drinking behavior proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-2 A method for intelligent monitoring of sheep weight based on drinking behavior includes the following steps: S1. Continuously monitor the weight sensor data of the drinking water terminal and trigger the drinking water monitoring state according to the preset threshold; S2. Under drinking water monitoring status, the radio frequency response signal in the drinking trough identification area is captured in time and the identifier is parsed to extract the valid electronic identifier of the current drinking process and form the sheep's identity result. S3. Perform time-series unfolding on the weight sensing data, and use the variable point detection method to segment the continuous interval of weight fluctuation, eliminate the unsteady load-bearing section caused by attitude change, and screen out the stable load-bearing section. S4. Perform sliding median filtering on the weight sampling values ​​within the stable bearing section, and use least squares fitting to construct the bearing trend line. Constrain the residual fluctuations that deviate from the bearing trend line to obtain the stable weight value. S5. Based on stable weight values, perform dynamic time warping and registration on the corresponding time intervals, and solve for the optimal registration path under the constraints of a preset matching window to generate individual binding results; S6. Perform structured packaging and link transmission organization on the individual binding results, and send the corresponding monitoring data to the remote server through the communication module; S7. Perform historical aggregation on the received monitoring data on the remote server, and use the moving average method to construct the individual weight evolution curve, identify abnormal offset segments, and output weight monitoring results and abnormal warning information.

[0023] In this embodiment, S1 specifically includes: S11. Perform continuous sampling on the weight sensing data output by the drinking water terminal to form a weight response sequence arranged in time order; S12. Perform continuous interval scanning on the weight response sequence, extract the static dwelling interval where the weight fluctuation amplitude is lower than the preset fluctuation threshold and the duration reaches the preset length, and solve the mean value of the weight sampled values ​​in the static dwelling interval to obtain the static bearing baseline as a reference for the weight of the drinking water terminal in the static state. S13. Perform difference expansion between each weight sample value in the weight response sequence and the static load baseline to obtain the weight offset at each time point, and construct a static threshold response based on the comparison result between the weight offset and the preset trigger increment threshold. S14. Perform continuous dwell confirmation on the static threshold response. When the weight offset continuously exceeds the preset static threshold and the continuous over-limit time reaches the preset duration threshold, the drinking water monitoring state is triggered. The preset static threshold is the weight trigger threshold formed by superimposing the preset trigger increment on the static bearing baseline. The drinking water monitoring state is the activation state in which the drinking water terminal switches from weight monitoring to identity recognition and subsequent weight measurement.

[0024] In this embodiment, S2 specifically includes: S21. Under drinking water monitoring status, control the radio frequency identification device in the drinking trough identification area to perform periodic gating excitation according to the preset polling rhythm, so that the electronic tag entering the identification area emits the corresponding radio frequency response in each polling cycle; S22. Perform time slot occupancy analysis on each radio frequency response along the polling cycle to determine the response time slot position of each electronic identifier in the current polling cycle, forming a candidate response set arranged in time slot order; S23. A time-slot anti-collision method is used to resolve conflicts in the candidate response set. Overlapping response time slots with concurrent occupancy are re-polled for allocation, and independent response time slots without concurrent occupancy are retained as candidate identifier time slots. Specifically, this includes: The current polling period is divided into multiple response time slots, and the response results corresponding to each electronic identifier are mapped to the corresponding response time slot according to the order in which each electronic identifier enters the radio frequency response in the candidate response set. Iterate through the number of occupants in each response time slot, and determine the response time slot in which two or more electronic tags are occupied at the same time as the conflict time slot. For conflicting time slots, perform time slot anti-collision processing, and reissue subsequent polling instructions to each electronic identifier in the conflicting time slot, so that each electronic identifier enters a different response time slot in the subsequent polling cycle according to the reassigned time slot sequence number; For each response time slot reallocated in subsequent polling cycles, placeholder detection is performed again. Response time slots with multiple electronic identifiers occupying the slot simultaneously are marked as conflict time slots, and time slot anti-collision processing is repeated until only a single electronic identifier occupies each reserved response time slot. Response time slots with only a single electronic identifier occupying the slot are determined as candidate identifier time slots. S24. Perform integrity checks on the response results within the candidate identifier time slots, remove abnormal response time slots corresponding to incomplete responses, interrupted responses, and abnormal occupancy, and retain valid response time slots that meet the continuous response conditions. S25. Perform continuous dominance locking on the electronic identifiers corresponding to valid response time slots, and determine the electronic identifiers that stably occupy positions within multiple consecutive polling cycles as the target electronic identifiers for the current drinking process, forming sheep identification results, specifically including: Extract the electronic identifiers corresponding to each valid response time slot along the continuous polling cycle, and construct the electronic identifier response sequence according to the polling order; The number of times each electronic identifier appears in the electronic identifier response sequence within the continuous polling period is accumulated, and the continuous occupancy length of the corresponding effective response time slot of each electronic identifier is simultaneously counted. Electronic identifiers that appear a number of times reaching a preset dominant frequency threshold and whose continuous occupancy length reaches a preset occupancy length threshold are identified as dominant candidate identifiers. When there are multiple dominant candidate identifiers, a stability comparison is performed on the continuous occupancy segments corresponding to each dominant candidate identifier, and the dominant candidate identifier with the smallest fluctuation in the continuous occupancy segment is retained as the target electronic identifier. The target electronic tag is identified as the identification tag corresponding to the current drinking process, thus forming the sheep identification result.

[0025] In this embodiment, S3 specifically includes: S31. The weight sensing data is unfolded according to the sampling time to form a weight fluctuation sequence, and the difference sequence between adjacent sampled values ​​is calculated along the weight fluctuation sequence. S32. The CUSUM change point detection method is used to perform cumulative offset search on the differential sequence. When the cumulative offset result exceeds the preset change threshold, the load change point is locked. The weight fluctuation sequence is divided into multiple continuous load segments according to each load change point. S33. Perform local slope fitting and segment range calculation on each continuous load-bearing section. Continuous load-bearing sections with local slope exceeding a preset slope threshold or segment range exceeding a preset fluctuation threshold are identified as unsteady load-bearing sections and are removed. The local slope fitting is to perform least squares fitting on the weight sampled values ​​in the continuous load-bearing section along the sampling time axis and extract the fitting slope. The segment range calculation is to calculate the difference between the maximum weight sampled value and the minimum weight sampled value in the continuous load-bearing section. S34. Perform a continuous length check on the remaining continuous load-bearing sections after stripping the unsteady load-bearing sections, and retain the continuous load-bearing sections whose continuous length reaches the preset length threshold as stable load-bearing sections.

[0026] In this embodiment, S32 specifically includes: S321. Perform positive and negative accumulation on each difference value in the difference sequence according to the sampling order to form a bidirectional cumulative offset sequence corresponding to each sampling time. S322. Perform cumulative offset search along the bidirectional cumulative offset sequence, and determine the sampling position where the absolute value of the cumulative offset continues to increase and exceeds the preset mutation threshold as the candidate mutation position. The preset mutation threshold is the cumulative offset judgment threshold that distinguishes between stable load changes and attitude mutation disturbances. S323. Using each candidate mutation location as the center, search forward along the sampling time axis to find the starting position of the cumulative offset, and search backward along the sampling time axis to find the falling position of the cumulative offset, and lock the sampling location where the absolute value of the cumulative offset reaches the extreme value as the carrying mutation point. S324. According to the sequential distribution of each load change point in the weight fluctuation sequence, the weight fluctuation sequence is divided into intervals, and the weight response interval between adjacent load change points is divided into continuous load segments.

[0027] In this embodiment, S4 specifically includes: S41. Along the time axis of the stable bearing section, the weight sample values ​​are normalized by sliding window midpoint. In each sliding window, the weight sample values ​​that deviate from the window midpoint and exceed the preset local fluctuation threshold are reset to the corresponding window midpoint, forming a normalized weight sequence with suppressed local disturbance. S42. Using the sampling time of the regular weight sequence as the independent variable and the weight value as the dependent variable, the least squares fitting is used to perform overall trend approximation on the stable bearing section, and the slope parameter and intercept parameter of the bearing trend line are solved to make the weight response in the regular weight sequence concentrate towards the bearing trend line. S43. Apply residual offset constraints to the regular weight sequence along the bearing trend line, establish offset tolerance boundaries on both sides of the bearing trend line, and apply convergence restrictions to weight sample values ​​exceeding the offset tolerance boundaries to form a constrained weight sequence, specifically including: A two-sided tolerance expansion is performed on the bearing trend line along the time axis of the stable bearing section. Constraint boundaries corresponding to the positive offset tolerance and the negative offset tolerance are constructed on both sides of the bearing trend line to form a two-sided tolerance constraint band surrounding the bearing trend line. Trend alignment is performed on the regular weight sequence along the double-sided tolerance constraint band, and the deviation of each weight sample value relative to the bearing trend line is projected onto the tolerance coordinates of the corresponding sampling time to form a residual offset sequence arranged in time sequence. Perform a two-sided over-limit scan on the residual offset sequence to separate the upper offset sampling points that cross the positive tolerance boundary and the lower offset sampling points that cross the reverse tolerance boundary, and mark each over-limit sampling point as exceeding the limit. The Huber constraint function is used to apply convergence constraints to the weight sample values ​​corresponding to the over-limit marker, compressing the upper offset sample points into the positive tolerance boundary and pulling the lower offset sample points into the reverse tolerance boundary, thus forming a boundary-constrained weight response. Sequence reconstruction is performed on the boundary-constrained weight response along the original sampling time sequence, so that the weight sampled values ​​after convergence constraint are re-embedded into the time position corresponding to the regular weight sequence to form a constrained weight sequence. S44. Perform convergence fitting confirmation on the restricted weight sequence. When the slope change and intercept change of two adjacent fitting results both fall within the preset convergence threshold, lock the convergence bearing trend line. S45. Perform stable weight center locking on the restricted weight sequence along the convergent bearing trend line, and determine the center value with the smallest total residual offset and the largest continuous residence length as the stable weight value.

[0028] In this embodiment, S5 specifically includes: S51. Extract the identification time series corresponding to the sheep's identity results and the measurement time series corresponding to the stable weight values, and apply time axis constraints to the identification time series and measurement time series according to the preset matching window to form the time series to be registered. S52. Construct a dynamic time warping cost matrix for the time series to be registered, and recursively accumulate the registration cost along the row and column directions of the cost matrix to form the temporal registration cost distribution between each identification time node and each measurement time node. S53. Under the constraint of a preset matching window, perform path search on the temporal registration cost distribution, block out-of-window paths that exceed the preset matching window, and find the optimal registration path with the minimum cumulative registration cost among the remaining feasible paths. Specifically, this includes: A feasible search band defined by a preset matching window is delineated along the temporal registration cost distribution, and cost units falling outside the feasible search band are locked as window-crossing units; Starting from the initial cost cell, the neighborhood of adjacent cost cells is expanded along the row, column and diagonal directions within the feasible search band to construct a candidate registration path formed by connecting adjacent cost cells end to end. For each candidate registration path, perform cumulative cost recursion, and update the cumulative cost of the current cost unit to the sum of the minimum cumulative cost of the preceding adjacent cost units and the registration cost of the current cost unit; Perform full path backtracking on the candidate registration paths that reach the termination cost unit, and lock the candidate registration path with the minimum cumulative path cost as the optimal registration path; S54. Map the identification time node and the measurement time node one-to-one along the optimal registration path, and perform association locking on the sheep identity results and stable weight values ​​that fall within the same path mapping unit to generate individual binding results.

[0029] In this embodiment, S6 specifically includes: S61. The individual binding results are structured and arranged according to the preset data organization order to form monitoring data units corresponding to the current drinking water process; S62. Perform frame encapsulation on the monitoring data unit, and add a frame header identifier and a check field to the outside of the monitoring data unit to form a data frame to be sent. The frame header identifier is a frame start identifier placed at the front of the monitoring data unit and used to mark the start boundary of the data frame. The check field is a check code field calculated based on each byte of data in the monitoring data unit and used for link transmission integrity verification. S63. Perform sending queue arrangement on the data frames to be sent, write each data frame to be sent into the sending queue in the order of sending, and assign a corresponding sending sequence number to each data frame to be sent. S64. The drive communication module sends each data frame to be sent sequentially according to the sending sequence number, and performs verification on the returned link response result. When the link response result indicates that the sending has failed, the corresponding data frame to be sent is retransmitted until the data transmission is completed. The communication module is used to send the data frame to be sent to the remote server and receive the link response result. The remote server is a server device set outside the drinking water terminal and establishing a data communication connection with the communication module.

[0030] In this embodiment, S7 specifically includes: S71. On the remote server, perform historical merging and time normalization on the time-series monitoring units along the monitoring time axis, and rearrange the time-series monitoring units corresponding to the same sheep into a continuous weight monitoring sequence according to the monitoring time. S72. Implement moving average recursion on the continuous weight monitoring sequence, perform mean update and sequence smoothing on the weight monitoring values ​​in each sliding window, and form the individual weight evolution curve corresponding to the monitoring time axis. S73. Perform offset trajectory expansion along the individual weight evolution curve, map the offset between the weight monitoring value at each monitoring time and the corresponding moving average to the monitoring time axis, and implement anomaly locking for monitoring segments with continuous offset directions and offset amplitudes that continuously exceed the preset offset threshold. S74. Perform type merging and result mapping on the monitoring segments after anomaly locking, organize the anomaly monitoring segments and individual weight evolution curves into weight monitoring results, and output the corresponding anomaly warning information.

[0031] refer to Figure 3 A smart weight monitoring system for sheep based on drinking behavior, comprising: The monitoring trigger module is used to continuously monitor the weight sensor data of the drinking water terminal and trigger the drinking water monitoring status according to the preset threshold. The identification module is used to capture and analyze the radio frequency response signal in the identification area of ​​the drinking trough in a timely manner during drinking water monitoring, extract the valid electronic identifier of the current drinking process, and form the sheep's identification result. The segment screening module is used to perform time-series unfolding of weight sensing data and to segment the continuous interval of weight fluctuation using a change point detection method, thereby eliminating unsteady load-bearing segments caused by attitude changes and screening out stable load-bearing segments. The weight calculation module is used to perform sliding median filtering on the weight sample values ​​within the stable bearing section, and to construct the bearing trend line using least squares fitting. It also constrains residual fluctuations that deviate from the bearing trend line to obtain stable weight values. The binding and registration module is used to perform dynamic time-warped registration of the corresponding time segment based on stable weight values, and to solve for the optimal registration path under the constraints of a preset matching window to generate individual binding results; The unit construction module is used to perform structured packaging and link transmission organization of individual binding results, and send the corresponding monitoring data to the remote server through the communication module; The analysis and early warning module is used to perform historical aggregation on the received monitoring data on a remote server, construct an individual weight evolution curve using the moving average method, identify abnormal offset segments, and output weight monitoring results and abnormal early warning information.

[0032] Example 1: To verify the feasibility of this invention in practice, it was applied to a routine individual weight monitoring scenario in a large-scale sheep farm. In this scenario, traditional weight acquisition mainly relies on manually driving sheep into a weighing device for centralized measurement. This is not only cumbersome and labor-intensive, but also prone to sheep hiding, struggling, and short-term agitation during the driving and weighing process, resulting in low weighing frequency, scattered data, and difficulty in timely grasping continuous changes in individual weight. Especially when sheep experience slow weight loss, stagnant growth, or short-term abnormal fluctuations, relying solely on low-frequency manual weighing often fails to detect these issues in time, affecting subsequent feeding adjustments and health assessments.

[0033] To solve the above problems, the intelligent monitoring system of this invention is deployed in the drinking area, integrating the drinking trough, weight sensor, radio frequency identification device and control terminal into one unit, so that the sheep can automatically complete weight monitoring during natural drinking.

[0034] During system operation, the weight sensor data of the drinking terminal is continuously monitored. When the weight deviation continuously exceeds the trigger threshold, the system enters the drinking monitoring state. When the sheep's head enters the identification area of ​​the drinking trough, electronic identification is completed. Then, the continuous weight data during the drinking process is unfolded over time. The non-steady-state bearing segment caused by head raising, head retraction, body deviation, and hoof movement is separated using the change point detection and stable bearing segment screening mechanism. Then, the stable bearing segment is subjected to moving median normalization, least squares fitting, and convergent weight center locking to obtain the stable weight value corresponding to a single drinking process. After that, individual binding results are generated through time segment registration and a time-series monitoring unit is formed along the monitoring time axis for subsequent continuous weight evolution analysis.

[0035] After one monitoring cycle, a total of 120 sheep were effectively identified, and a total of 6,840 drinking triggers were recorded. Among them, 6,128 valid weighing records successfully formed individual binding results, accounting for 89.59% of the total. Each sheep obtained an average of 51.07 valid weight records, which is significantly higher than the average of 4 records per sheep under the manual discrete weighing method.

[0036] After manual static weighing verification, the average absolute error between the stable weight value obtained by this invention and the manual weighing result was 0.42 kg, and the proportion of records with errors controlled within 0.80 kg reached 93.10%. At the same time, the average standard deviation of weight fluctuation in the original drinking water weighing data was 1.36 kg. After screening the stable bearing section and solving the stable weight value, the standard deviation of fluctuation within the stable section decreased to 0.31 kg, indicating that this invention can effectively suppress instantaneous disturbances caused by posture changes.

[0037] During continuous monitoring, the system further performed a moving average analysis on the individual weight evolution curves and identified abnormal deviation segments. Monitoring results showed that 18 sheep were identified as having abnormal deviation segments, of which 12 exhibited a continuous downward deviation and 6 exhibited a stagnant growth deviation. Subsequent feeding observation revealed that 15 of these sheep did indeed show reduced feed intake, fluctuating mental state, or abnormal watering frequency, achieving an abnormal identification accuracy rate of 83.33%.

[0038] Overall, this invention balances high monitoring frequency, good stability of weight measurement, and strong anomaly identification capability, effectively improving the problems of high stress interference, low acquisition frequency, and insufficient continuous individual tracking in traditional sheep weight monitoring.

[0039] Table 1. Statistical Table of the Implementation Effect of Intelligent Monitoring of Sheep Weight

[0040] As can be seen from Table 1, the present invention outperforms the comparison method in terms of the number of effective weighing records, the average number of records per sheep, the average absolute error, and the anomaly identification accuracy rate. This indicates that the present invention can achieve higher frequency, more stable, and more continuous intelligent weight monitoring during the natural drinking process of sheep.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of sheep weight based on drinking behavior, characterized in that, Includes the following steps: S1. Continuously monitor the weight sensor data of the drinking water terminal and trigger the drinking water monitoring state according to the preset threshold; S2. Under drinking water monitoring status, the radio frequency response signal in the drinking trough identification area is captured in time and the identifier is parsed to extract the valid electronic identifier of the current drinking process and form the sheep's identity result. S3. Perform time-series unfolding on the weight sensing data, and use the variable point detection method to segment the continuous interval of weight fluctuation, eliminate the unsteady load-bearing section caused by attitude change, and screen out the stable load-bearing section. S4. Perform sliding median filtering on the weight sampling values ​​within the stable bearing section, and use least squares fitting to construct the bearing trend line. Constrain the residual fluctuations that deviate from the bearing trend line to obtain the stable weight value. S5. Based on stable weight values, perform dynamic time warping and registration on the corresponding time intervals, and solve for the optimal registration path under the constraints of a preset matching window to generate individual binding results; S6. Perform structured packaging and link transmission organization on the individual binding results, and send the corresponding monitoring data to the remote server through the communication module; S7. Perform historical aggregation on the received monitoring data on the remote server, and use the moving average method to construct the individual weight evolution curve, identify abnormal offset segments, and output weight monitoring results and abnormal warning information.

2. The intelligent weight monitoring method for sheep based on drinking behavior according to claim 1, characterized in that, S1 specifically includes: S11. Perform continuous sampling on the weight sensing data output by the drinking water terminal to form a weight response sequence arranged in time order; S12. Perform continuous interval scanning on the weight response sequence, extract the static dwelling interval where the weight fluctuation amplitude is lower than the preset fluctuation threshold and the duration reaches the preset length, and solve the mean value of the weight sampled values ​​in the static dwelling interval to obtain the static bearing baseline as a reference for the weight of the drinking water terminal in the static state. S13. Perform difference expansion between each weight sample value in the weight response sequence and the static load baseline to obtain the weight offset at each time point, and construct a static threshold response based on the comparison result between the weight offset and the preset trigger increment threshold. S14. Perform continuous dwell confirmation on the static threshold response. When the weight offset continuously exceeds the preset static threshold and the continuous over-limit time reaches the preset duration threshold, the drinking water monitoring state is triggered. The preset static threshold is the weight trigger threshold formed by superimposing the preset trigger increment on the static bearing baseline. The drinking water monitoring state is the activation state in which the drinking water terminal switches from weight monitoring to identity recognition and subsequent weight measurement.

3. The intelligent weight monitoring method for sheep based on drinking behavior according to claim 1, characterized in that, S2 specifically includes: S21. Under drinking water monitoring status, control the radio frequency identification device in the drinking trough identification area to perform periodic gating excitation according to the preset polling rhythm, so that the electronic tag entering the identification area emits the corresponding radio frequency response in each polling cycle; S22. Perform time slot occupancy analysis on each radio frequency response along the polling cycle to determine the response time slot position of each electronic identifier in the current polling cycle, forming a candidate response set arranged in time slot order; S23. Use a time slot anti-collision method to resolve conflicts in the candidate response set, re-polling and allocating overlapping response time slots with concurrent occupancy, and retaining independent response time slots without concurrent occupancy as candidate identifier time slots. S24. Perform integrity checks on the response results within the candidate identifier time slots, remove abnormal response time slots corresponding to incomplete responses, interrupted responses, and abnormal occupancy, and retain valid response time slots that meet the continuous response conditions. S25. Perform continuous dominance locking on the electronic identifiers corresponding to valid response time slots, and determine the electronic identifiers that stably occupy positions within multiple consecutive polling cycles as the target electronic identifiers for the current drinking process, forming sheep identification results, specifically including: Extract the electronic identifiers corresponding to each valid response time slot along the continuous polling cycle, and construct the electronic identifier response sequence according to the polling order; The number of times each electronic identifier appears in the electronic identifier response sequence within the continuous polling period is accumulated, and the continuous occupancy length of the corresponding effective response time slot of each electronic identifier is simultaneously counted. Electronic identifiers that appear a number of times reaching a preset dominant frequency threshold and whose continuous occupancy length reaches a preset occupancy length threshold are identified as dominant candidate identifiers. When there are multiple dominant candidate identifiers, a stability comparison is performed on the continuous occupancy segments corresponding to each dominant candidate identifier, and the dominant candidate identifier with the smallest fluctuation in the continuous occupancy segment is retained as the target electronic identifier. The target electronic tag is identified as the identification tag corresponding to the current drinking process, thus forming the sheep identification result.

4. The intelligent weight monitoring method for sheep based on drinking behavior according to claim 1, characterized in that, S3 specifically includes: S31. The weight sensing data is unfolded according to the sampling time to form a weight fluctuation sequence, and the difference sequence between adjacent sampled values ​​is calculated along the weight fluctuation sequence. S32. The CUSUM change point detection method is used to perform cumulative offset search on the differential sequence. When the cumulative offset result exceeds the preset change threshold, the load change point is locked. The weight fluctuation sequence is divided into multiple continuous load segments according to each load change point. S33. Perform local slope fitting and segment range calculation on each continuous load-bearing section. Continuous load-bearing sections with local slope exceeding a preset slope threshold or segment range exceeding a preset fluctuation threshold are identified as unsteady load-bearing sections and are removed. The local slope fitting is to perform least squares fitting on the weight sampled values ​​in the continuous load-bearing section along the sampling time axis and extract the fitting slope. The segment range calculation is to calculate the difference between the maximum weight sampled value and the minimum weight sampled value in the continuous load-bearing section. S34. Perform a continuous length check on the remaining continuous load-bearing sections after stripping the unsteady load-bearing sections, and retain the continuous load-bearing sections whose continuous length reaches the preset length threshold as stable load-bearing sections.

5. The intelligent weight monitoring method for sheep based on drinking behavior according to claim 4, characterized in that, Specifically, S32 includes: S321. Perform positive and negative accumulation on each difference value in the difference sequence according to the sampling order to form a bidirectional cumulative offset sequence corresponding to each sampling time. S322. Perform cumulative offset search along the bidirectional cumulative offset sequence, and determine the sampling position where the absolute value of the cumulative offset continues to increase and exceeds the preset mutation threshold as the candidate mutation position. The preset mutation threshold is the cumulative offset judgment threshold that distinguishes between stable load changes and attitude mutation disturbances. S323. Using each candidate mutation location as the center, search forward along the sampling time axis to find the starting position of the cumulative offset, and search backward along the sampling time axis to find the falling position of the cumulative offset, and lock the sampling location where the absolute value of the cumulative offset reaches the extreme value as the carrying mutation point. S324. According to the sequential distribution of each load change point in the weight fluctuation sequence, the weight fluctuation sequence is divided into intervals, and the weight response interval between adjacent load change points is divided into continuous load segments.

6. The intelligent weight monitoring method for sheep based on drinking behavior according to claim 1, characterized in that, S4 specifically includes: S41. Along the time axis of the stable bearing section, the weight sample values ​​are normalized by sliding window midpoint. In each sliding window, the weight sample values ​​that deviate from the window midpoint and exceed the preset local fluctuation threshold are reset to the corresponding window midpoint, forming a normalized weight sequence with suppressed local disturbance. S42. Using the sampling time of the regular weight sequence as the independent variable and the weight value as the dependent variable, the least squares fitting is used to perform overall trend approximation on the stable bearing section, and the slope parameter and intercept parameter of the bearing trend line are solved to make the weight response in the regular weight sequence concentrate towards the bearing trend line. S43. Apply residual offset constraints to the regular weight sequence along the bearing trend line, establish offset tolerance boundaries on both sides of the bearing trend line, and apply convergence restrictions to weight sample values ​​exceeding the offset tolerance boundaries to form a constrained weight sequence, specifically including: A two-sided tolerance expansion is performed on the bearing trend line along the time axis of the stable bearing section. Constraint boundaries corresponding to the positive offset tolerance and the negative offset tolerance are constructed on both sides of the bearing trend line to form a two-sided tolerance constraint band surrounding the bearing trend line. Trend alignment is performed on the regular weight sequence along the double-sided tolerance constraint band, and the deviation of each weight sample value relative to the bearing trend line is projected onto the tolerance coordinates of the corresponding sampling time to form a residual offset sequence arranged in time sequence. Perform a two-sided over-limit scan on the residual offset sequence to separate the upper offset sampling points that cross the positive tolerance boundary and the lower offset sampling points that cross the reverse tolerance boundary, and mark each over-limit sampling point as exceeding the limit. The Huber constraint function is used to apply convergence constraints to the weight sample values ​​corresponding to the over-limit marker, compressing the upper offset sample points into the positive tolerance boundary and pulling the lower offset sample points into the reverse tolerance boundary, thus forming a boundary-constrained weight response. Sequence reconstruction is performed on the boundary-constrained weight response along the original sampling time sequence, so that the weight sampled values ​​after convergence constraint are re-embedded into the time position corresponding to the regular weight sequence to form a constrained weight sequence. S44. Perform convergence fitting confirmation on the restricted weight sequence. When the slope change and intercept change of two adjacent fitting results both fall within the preset convergence threshold, lock the convergence bearing trend line. S45. Perform stable weight center locking on the restricted weight sequence along the convergent bearing trend line, and determine the center value with the smallest total residual offset and the largest continuous residence length as the stable weight value.

7. The intelligent weight monitoring method for sheep based on drinking behavior according to claim 1, characterized in that, S5 specifically includes: S51. Extract the identification time series corresponding to the sheep's identity results and the measurement time series corresponding to the stable weight values, and apply time axis constraints to the identification time series and measurement time series according to the preset matching window to form the time series to be registered. S52. Construct a dynamic time warping cost matrix for the time series to be registered, and recursively accumulate the registration cost along the row and column directions of the cost matrix to form the temporal registration cost distribution between each identification time node and each measurement time node. S53. Under the constraints of the preset matching window, perform path search on the temporal registration cost distribution, block out-of-window paths that exceed the preset matching window, and find the optimal registration path with the minimum cumulative registration cost among the remaining feasible paths. S54. Map the identification time node and the measurement time node one-to-one along the optimal registration path, and perform association locking on the sheep identity results and stable weight values ​​that fall within the same path mapping unit to generate individual binding results.

8. The intelligent monitoring method for sheep weight based on drinking behavior according to claim 1, characterized in that, S6 specifically includes: S61. The individual binding results are structured and arranged according to the preset data organization order to form monitoring data units corresponding to the current drinking water process; S62. Perform frame encapsulation on the monitoring data unit, and add a frame header identifier and a check field to the outside of the monitoring data unit to form a data frame to be sent. S63. Perform sending queue arrangement on the data frames to be sent, write each data frame to be sent into the sending queue in the order of sending, and assign a corresponding sending sequence number to each data frame to be sent. S64. The drive communication module sends each data frame to be sent sequentially according to the sending sequence number, and performs verification on the returned link response result. When the link response result indicates that the sending has failed, the corresponding data frame to be sent is retransmitted until the data transmission is completed. The communication module is used to send the data frame to be sent to the remote server and receive the link response result. The remote server is a server device set outside the drinking water terminal and establishing a data communication connection with the communication module.

9. The intelligent monitoring method for sheep weight based on drinking behavior according to claim 1, characterized in that, Specifically, S7 includes: S71. On the remote server, perform historical merging and time normalization on the time-series monitoring units along the monitoring time axis, and rearrange the time-series monitoring units corresponding to the same sheep into a continuous weight monitoring sequence according to the monitoring time. S72. Implement moving average recursion on the continuous weight monitoring sequence, perform mean update and sequence smoothing on the weight monitoring values ​​in each sliding window, and form the individual weight evolution curve corresponding to the monitoring time axis. S73. Perform offset trajectory expansion along the individual weight evolution curve, map the offset between the weight monitoring value at each monitoring time and the corresponding moving average to the monitoring time axis, and implement anomaly locking for monitoring segments with continuous offset directions and offset amplitudes that continuously exceed the preset offset threshold. S74. Perform type merging and result mapping on the monitoring segments after anomaly locking, organize the anomaly monitoring segments and individual weight evolution curves into weight monitoring results, and output the corresponding anomaly warning information.

10. A sheep weight intelligent monitoring system based on drinking behavior, comprising the sheep weight intelligent monitoring method based on drinking behavior as described in any one of claims 1 to 9, characterized in that, include: The monitoring trigger module is used to continuously monitor the weight sensor data of the drinking water terminal and trigger the drinking water monitoring status according to the preset threshold. The identification module is used to capture and analyze the radio frequency response signal in the identification area of ​​the drinking trough in a timely manner during drinking water monitoring, extract the valid electronic identifier of the current drinking process, and form the sheep's identification result. The segment screening module is used to perform time-series unfolding of weight sensing data and to segment the continuous interval of weight fluctuation using a change point detection method, thereby eliminating unsteady load-bearing segments caused by attitude changes and screening out stable load-bearing segments. The weight calculation module is used to perform sliding median filtering on the weight sample values ​​within the stable bearing section, and to construct the bearing trend line using least squares fitting. It also constrains residual fluctuations that deviate from the bearing trend line to obtain stable weight values. The binding and registration module is used to perform dynamic time-warped registration of the corresponding time segment based on stable weight values, and to solve for the optimal registration path under the constraints of a preset matching window to generate individual binding results; The unit construction module is used to perform structured packaging and link transmission organization of individual binding results, and send the corresponding monitoring data to the remote server through the communication module; The analysis and early warning module is used to perform historical aggregation on the received monitoring data on a remote server, construct an individual weight evolution curve using the moving average method, identify abnormal offset segments, and output weight monitoring results and abnormal early warning information.

Images