A method for intelligent analysis of traditional Chinese medicine pulse diagnosis data based on pattern recognition

Abstract

This invention discloses an intelligent analysis method for TCM pulse diagnosis data based on pattern recognition, comprising the following steps: collecting and aligning pulse waves of the left and right hands' cun, guan, and chi positions using the superficial-middle-deep palpation method to form a layered pulse position sampling sequence; performing periodic positioning and five-phase division on the sampling sequence to form a pulse phase sequence; closing the beginning and end of the single-beat cycle of each pulse position and level to form a set of closed-loop pulse paths; performing a logarithmic signature transformation of the pulse constraint quotient on the closed-loop paths to form a set of pulse structure spectra; calculating the single-beat closed-loop deviation, pulse position exchange sub-spectrum, and closed-loop residual by combining the closed-loop paths and structure spectra; and completing pulse identification and outputting structural interpretation results based on the structure spectrum, pulse position exchange sub-spectrum, and closed-loop residual. This invention uses the logarithmic signature method of the pulse constraint quotient to achieve structured analysis and identification of TCM pulse diagnosis data, which has the advantages of accurate identification, stable results, and strong interpretability.

Description

Technical Field

[0001] This invention relates to the field of information technology for TCM diagnosis, and in particular to an intelligent analysis method for TCM pulse diagnosis data based on pattern recognition. Background Technology

[0002] Pulse diagnosis is a crucial basis for TCM syndrome differentiation. Current digital pulse diagnosis technologies typically use pressure sensors, pulse wave acquisition devices, or multi-channel detection terminals to collect pulse wave signals at different pulse-taking levels at the cun, guan, and chi positions on both hands. These signals are then processed using methods such as filtering, noise reduction, period segmentation, peak-valley detection, and statistical parameter extraction to analyze changes in pulse position, pulse rate, pulse strength, amplitude, pulse width, and rhythm. Rule-based judgment or machine learning models are then used to identify pulse types. These methods have established a certain foundation in the objective acquisition and preliminary quantification of pulse waves, providing technical support for the standardization, digitization, and intelligentization of TCM pulse diagnosis.

[0003] However, most existing technologies still focus on analyzing single-point parameters, local waveform indicators, or overall statistics of pulse waves. They lack a detailed expression of the sequential relationship between the onset, rising, peak, falling, and residual momentum within a single pulse cycle. They also lack unified modeling of the coupling relationship between adjacent pulse positions on the same hand, corresponding pulse positions on the left and right, different pulse-taking levels, and adjacent single pulse cycles. This results in insufficient utilization of pulse structure information, making it difficult to simultaneously characterize the evolution within the pulse position, the synergy between pulse positions, and the cross-cycle inheritance characteristics. Consequently, the interpretability of pulse recognition results is insufficient, and the source of structural deviations is difficult to trace, affecting the accuracy and stability of intelligent analysis of complex pulses.

[0004] Therefore, how to provide a pattern recognition-based intelligent analysis method for TCM pulse diagnosis data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent analysis method for TCM pulse diagnosis data based on pattern recognition. This invention uses the pulse constrained quotient logarithmic signature method to realize the structured analysis and recognition of TCM pulse diagnosis data, which has the advantages of accurate recognition, stable results and strong interpretability.

[0006] According to an embodiment of the present invention, a method for intelligent analysis of TCM pulse diagnosis data based on pattern recognition includes the following steps: Acquire pulse wave data of the left and right hands at the Cun, Guan, and Chi positions under the floating, middle, and deep positions, align them according to the pulse position identifier, level identifier, and time index, and generate a pulse position hierarchical sampling sequence. Perform pulsation cycle localization and phase division on the pulse position layered sampling sequence to generate a pulse phase sequence containing the onset segment, rising segment, peak segment, falling segment and residual segment; Based on the pulse phase sequence, the single beat cycle of each pulse position and level is closed at the beginning and end while maintaining the phase order, generating a set of closed-loop pulse paths. Perform a pulse constraint quotient logarithmic signature transformation on the pulse closed-loop path set to extract intra-pulse phase exchange terms, inter-pulse exchange terms, hierarchical migration terms, and cross-period inheritance terms to generate a pulse structure spectrum set. The single-beat closed-loop deviation is calculated based on the pulse closed-loop path set, and the pulse exchange sub-spectrum of adjacent pulse positions in the same hand, corresponding pulse positions on the left and right and cross-level pulse positions is calculated based on the pulse structure spectrum set. The closed-loop residual result is generated by combining the single-beat closed-loop deviation. Pulse image recognition is performed based on the pulse image structure spectrum set, pulse position exchange sub-spectrum, and closed-loop residual results. The output pulse image category results and structure interpretation results are provided. The structure interpretation results include the corresponding dominant exchange sub-items, pulse position imbalance items, hierarchical migration items, and cross-cycle inheritance items.

[0007] Optionally, the generation of the pulse position hierarchical sampling sequence specifically includes: Pulse wave data, sampling timestamps, and pulse pressure values ​​were collected at the left hand cun, left hand guan, left hand chi, right hand cun, right hand guan, and right hand chi positions respectively, under the conditions of superficial, middle, and deep sampling, to form the original sampling set for pulse diagnosis; Perform pulse diagnosis hierarchical segmentation and stable segment extraction on the original pulse diagnosis sampling set to obtain a set of hierarchical effective sampling segments; Anomaly removal and baseline trimming are performed on the hierarchical effective sampled segment set to obtain a purified pulse wave segment set; Perform uniform sampling interval resampling and time index reconstruction on the purified pulse wave fragment set to obtain a time-aligned sample set; Based on the hand-side position, pulse position, pulse retrieval level, and unified time index, the time-aligned sampling set is subjected to identifier binding and sequential encapsulation to obtain the identifier-aligned sampling set; Amplitude normalization is performed on the identifier-aligned sampling set, and the results are output according to the pulse position identifier, level identifier, and unified time index to generate a pulse position hierarchical sampling sequence.

[0008] Optionally, the generation of the pulse phase sequence specifically includes: Local smoothing and waveform extremum extraction are performed on the pulse position hierarchical sampling sequence grouped by pulse position identifier and hierarchical identifier to obtain a set of periodic candidate labels; The pulsation cycle is located by performing the correspondence between adjacent valley points and the main peak in the candidate periodic marker set, thus forming a set of single-beat cycles; For a single-wave cycle set, the starting inflection point, the rising end point, the peak end point, and the falling end point are extracted to obtain a set of phase boundary results. Phase segmentation is performed on each single-beat cycle based on the phase boundary result set to obtain the phase segmentation result set; Perform phase order verification and abnormal phase removal on the phase segmentation result set to form a valid phase segment set; The effective phase segment set is sequentially encapsulated according to the single beat cycle identifier, pulse position identifier, level identifier and phase order to generate a pulse phase sequence.

[0009] Optionally, the generation of the pulse closed-loop path set specifically includes: The pulse phase sequence is grouped and organized according to the single-beat cycle identifier, pulse position identifier, and level identifier. The starting segment, rising segment, peak segment, falling segment, and residual segment within the same single-beat cycle are extracted to form a closed-loop input set. Connect the starting segment, rising segment, peak segment, falling segment and residual segment in sequence according to the phase order in the closed-loop construction input set to obtain the phase connection result set; Perform a head-and-tail closure construction on the phase connection result set to obtain a head-and-tail closure candidate set; Perform closure validity checks and sequence consistency checks on the candidate sets of closed paths at both ends to form a set of closed valid paths; Perform path index writing and phase mapping encapsulation on the closed valid path set to obtain the closed-loop path index set; The closed-loop path index set is sequentially encapsulated according to the pulse position identifier, level identifier, and single beat cycle identifier to generate the pulse closed-loop path set.

[0010] Optionally, the generation of the pulse structure spectrum set specifically includes: Extract the phase segment boundary position, closed connection segment, pulse position identifier, level identifier and single beat cycle identifier from the pulse closed loop path set, and construct a pulse constraint index table according to the pairing of adjacent phase segments within the same pulse position identifier, pairing of adjacent pulse positions in the same hand, pairing of corresponding pulse positions on the left and right, pairing of the same pulse position identifier across levels, and pairing of adjacent single beat cycles under the same level of the same pulse position identifier. According to the pulse constraint index table, perform path segmentation and path alignment on the pulse closed-loop path set to obtain the quotient logarithmic signature input set. Perform equivalent merging and repetition elimination on the input set of quotient logarithmic signatures to obtain the intermediate set of quotient logarithmic signatures; Perform pulse-constrained quotient logarithmic signature transformation on the intermediate set of quotient logarithmic signatures to obtain the set of intra-pulse phase exchange terms, the set of inter-pulse exchange terms, the set of hierarchical migration terms, and the set of cross-cycle inheritance terms. The set of phase exchange terms within pulse positions, the set of exchange terms between pulse positions, the set of hierarchical migration terms, and the set of cross-cycle inheritance terms are written with term-level identifiers to obtain the set of pulse structure spectrum entries. The pulse structure spectrum entries are uniformly encapsulated according to item type, pulse position identifier, level identifier, and single beat cycle identifier to generate a pulse structure spectrum set.

[0011] Optionally, the pulse constrained quotient logarithmic signature transformation reorganizes the path block pairing results corresponding to each merging unit in the intermediate set of quotient logarithmic signatures according to a unified time index order to obtain the local path of the merging unit. An ordered integral combination within a preset order is calculated for the local path of the merging unit, and a logarithmic mapping is applied to convert it into a local logarithmic signature representation within a preset order. The corresponding coordinates in the local logarithmic signature representation are selectively read according to the pairing type corresponding to the merging unit. Among them, reading the pairing of adjacent phase segments within the same pulse position identifier to obtain the phase exchange coordinates within the pulse position, reading the pairing of adjacent pulse positions on the same hand and the pairing of corresponding pulse positions on the left and right to obtain the exchange coordinates between pulse positions, reading the pairing of cross-levels of the same pulse position identifier to obtain the level migration coordinates, and reading the pairing of adjacent single beat cycles under the same level of the same pulse position identifier to obtain the cross-cycle inheritance coordinates.

[0012] Optionally, the generation of the closed-loop residual results specifically includes: Extract the closed loop segments, first sampling point, last sampling point, phase segment boundary position, pulse position identifier, hierarchical identifier and single beat cycle identifier from the pulse closed loop path set, and extract the intra-pulse phase exchange term, inter-pulse exchange term, hierarchical migration term and cross-cycle inheritance term from the pulse structure spectrum set to form the residual calculation input set; Perform head-and-tail closure deviation calculation and phase boundary deviation calculation on the deviation calculation input set to obtain the single-shot closed-loop deviation; Extract inter-pulse position exchange terms and hierarchical migration terms from the pulse structure spectrum set, perform grouping and aggregation according to pairing type, and obtain the exchange sub-spectrum input set; Perform directional pairing, reverse cancellation, and spectral encapsulation on the input set of exchanged sub-spectrums to obtain the pulse position exchanged sub-spectrums; The single-beat closed-loop deviation, pulse position exchange sub-spectrum, pulse position phase exchange term and cross-cycle inheritance term are jointly merged according to the same pulse position identifier, the same level identifier and the same single-beat cycle identifier to generate path deviation decomposition results; Based on the path deviation decomposition results, residual adjudication and residual classification are performed to generate closed-loop residual results.

[0013] Optionally, the residual determination is based on whether the single-beat closed-loop deviation exceeds a preset closure deviation threshold, whether the phase exchange residual component within the pulse position exceeds a preset phase exchange residual threshold, whether the exchange residual components of adjacent pulse positions on the same hand and the exchange residual components of corresponding pulse positions on the left and right exceed a preset pulse position exchange residual threshold, whether the cross-level pulse position exchange residual component exceeds a preset level migration residual threshold, and whether the cross-cycle inheritance residual component exceeds a preset inheritance residual threshold. Specifically, when the single-beat closed-loop deviation does not exceed the preset closure deviation threshold and each exchange residual component does not exceed the corresponding threshold, the decision is a complete closed-loop state. When the single-beat closed-loop deviation exceeds the preset closure deviation threshold and each exchange residual component does not exceed the corresponding threshold, the decision is a tense closed-loop state. When the single-beat closed-loop deviation does not exceed the preset closure deviation threshold and at least one exchange residual component exceeds the corresponding threshold, the decision is a deviated closed-loop state. When the single-beat closed-loop deviation exceeds the preset closure deviation threshold and at least two exchange residual components exceed the corresponding threshold, the decision is a broken closed-loop state. The closed-loop state obtained from the decision, the dominant residual source item, the corresponding pulse position identifier, the corresponding level identifier, and the corresponding single-beat cycle identifier are all written into the closed-loop residual result.

[0014] Optionally, the generation of the pulse category results and structural interpretation results specifically includes: Extract the intra-pulse phase exchange term, inter-pulse exchange term, hierarchical transfer term, and cross-cycle inheritance term from the pulse structure spectrum set. Extract the same-hand adjacent pulse exchange sub-spectrum, left-right corresponding pulse exchange sub-spectrum, and cross-hierarchical pulse exchange sub-spectrum from the pulse exchange sub-spectrum. Extract the closed-loop state and dominant residual source term from the closed-loop residual result to form the recognition input set. Based on the input set, perform category prototype matching to obtain a candidate set of pulse category; Perform category strength aggregation and category sorting on the candidate set of pulse categories to obtain a set of category score results; Based on the identification input set and the category scoring result set, dominant exchange sub-items, pulse position imbalance items, hierarchical transfer items and cross-period inheritance items are extracted to generate a candidate set of structural interpretations; A consistency check and result adjudication are performed on the category scoring result set and the structural interpretation candidate set to obtain the pulse recognition result set. Output the pulse category result and the structural interpretation result according to the adjudication results in the pulse recognition result set.

[0015] The beneficial effects of this invention are: This invention unifies and aligns pulse wave data from the Cun, Guan, and Chi positions of both hands under superficial, middle, and deep palpation techniques. This further completes the localization of the pulsation cycle, the division of the five phases, and the closure of the beginning and end of the single-beat cycle. It transforms the changes in the initial, rising, peak, falling, and residual momentum—processes difficult to quantify directly in traditional pulse diagnosis—into a set of closed-loop pulse path patterns with clear temporal boundaries and closed-loop structures. Based on this, it utilizes the logarithmic signature transformation of the pulse constraint quotient to extract phase exchange terms within pulse positions, exchange terms between pulse positions, hierarchical migration terms, and cross-cycle inheritance terms. This allows pulse diagnosis data to move beyond superficial parameter analysis such as amplitude, pulse rate, or local peaks and troughs, enabling a unified characterization of the sequential relationships within the single-beat cycle, the coupling relationships between adjacent pulse positions on the same hand and their corresponding left and right hands, the migration relationships between different pulse-taking levels, and the continuity relationships between adjacent single-beat cycles. This significantly enhances the structural expressive power of pulse information.

[0016] Furthermore, this invention, by constructing a pulse position exchange sub-spectrum and combining it with the single-beat closed-loop deviation to generate closed-loop residual results, can refine pulse deviations into beginning-and-end closure deviations, phase boundary deviations, intra-pulse position phase exchange residuals, inter-pulse position exchange residuals, cross-level migration residuals, and cross-cycle inheritance residuals, achieving hierarchical identification and directional localization of pulse abnormality sources. Based on the pulse structure spectrum set, pulse position exchange sub-spectrum, and closed-loop residual results, pulse identification is jointly performed, which not only outputs pulse category results but also simultaneously outputs structural interpretation results composed of dominant exchange sub-items, pulse position imbalance items, hierarchical migration items, and cross-cycle inheritance items. This improves the accuracy, stability, and interpretability of pulse identification, enhances the identification ability in complex and composite pulse scenarios, and makes pulse diagnosis results more consistent with the requirements of comprehensive judgment of position, potential, hierarchy, and continuous evolution relationships in traditional Chinese medicine. Attached Figure Description

[0017] 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 a flowchart of a method for intelligent analysis of TCM pulse diagnosis data based on pattern recognition, as proposed in this invention. Figure 2 This is a flowchart illustrating the generation of pulse structure spectrum in a pattern recognition-based intelligent analysis method for TCM pulse diagnosis data proposed in this invention. Figure 3 This is a flowchart illustrating the pulse pattern recognition and structural interpretation process of a pattern recognition-based intelligent analysis method for TCM pulse diagnosis data proposed in this invention. Detailed Implementation

[0018] 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.

[0019] refer to Figures 1-3 A method for intelligent analysis of TCM pulse diagnosis data based on pattern recognition includes the following steps: Acquire pulse wave data of the left and right hands at the Cun, Guan, and Chi positions under the floating, middle, and deep positions, align them according to the pulse position identifier, level identifier, and time index, and generate a pulse position hierarchical sampling sequence. Perform pulsation cycle localization and phase division on the pulse position layered sampling sequence to generate a pulse phase sequence containing the onset segment, rising segment, peak segment, falling segment and residual segment; Based on the pulse phase sequence, the single beat cycle of each pulse position and level is closed at the beginning and end while maintaining the phase order, generating a set of closed-loop pulse paths. Perform a pulse constraint quotient logarithmic signature transformation on the pulse closed-loop path set to extract intra-pulse phase exchange terms, inter-pulse exchange terms, hierarchical migration terms, and cross-period inheritance terms to generate a pulse structure spectrum set. The single-beat closed-loop deviation is calculated based on the pulse closed-loop path set, and the pulse exchange sub-spectrum of adjacent pulse positions in the same hand, corresponding pulse positions on the left and right and cross-level pulse positions is calculated based on the pulse structure spectrum set. The closed-loop residual result is generated by combining the single-beat closed-loop deviation. Pulse identification is performed based on the pulse structure spectrum set, pulse position exchange sub-spectrum, and closed-loop residual results. The output pulse category results and structure interpretation results are provided. The structure interpretation results include the corresponding dominant exchange sub-items, pulse position imbalance items, hierarchical migration items, and cross-cycle inheritance items.

[0020] In this embodiment, the generation of the pulse position hierarchical sampling sequence specifically includes: Pulse wave data, sampling timestamps, and pulse pressure values ​​were collected at the left hand cun, left hand guan, left hand chi, right hand cun, right hand guan, and right hand chi positions respectively, under the conditions of superficial, middle, and deep sampling, to form the original sampling set for pulse diagnosis; Each sampling record in the original pulse diagnosis sampling set includes the hand position, pulse position, pulse level, pulse wave data, sampling timestamp, and pulse pressure value. The hand position is distinguished between the left and right hands, the pulse position is distinguished between cun, guan, and chi, and the pulse level is distinguished between superficial, middle, and deep pulse. Perform pulse diagnosis hierarchical segmentation and stable segment extraction on the original pulse diagnosis sampling set to obtain a set of hierarchical effective sampling segments; The pulse taking level is divided into superficial, middle and deep segments according to the continuous change trajectory of pulse taking pressure value under the same hand side position and the same pulse position. The stable segment is extracted by removing pulse taking switching segments, pressure jump segments and sampling segments with insufficient duration within each pulse taking level, and the retained segments are written into the set of valid sampling segments of the level. Anomaly removal and baseline trimming are performed on the hierarchical effective sampled segment set to obtain a purified pulse wave segment set; The anomaly removal process generates anomaly markers for amplitude saturation, sampling gaps, instantaneous spikes, and periodic collapses, and deletes sampling points covered by the anomaly markers. The baseline trimming process performs local mean regression on the retained sampling points to remove slow drift and reconstructs them into a set of purified pulse wave segments according to the original time sequence. Perform uniform sampling interval resampling and time index reconstruction on the purified pulse wave fragment set to obtain a time-aligned sample set; Based on the hand-side position, pulse position, pulse retrieval level, and unified time index, the time-aligned sampling set is subjected to identifier binding and sequential encapsulation to obtain the identifier-aligned sampling set; The identifier binding combines the hand position and the pulse position to generate a pulse position identifier, writes the pulse taking level into the level identifier, and encapsulates the sampling points under the same pulse position identifier and the same level identifier, which are arranged in ascending order according to the same time index, into an identifier-aligned sampling set. Amplitude normalization is performed on the identifier-aligned sampling set, and the results are output according to pulse position identifier, level identifier, and unified time index to generate a pulse position hierarchical sampling sequence. The amplitude normalization method involves subtracting the median value of the same identifier-aligned sampling set from the original pulse wave value in the identifier-aligned sampling set, and then dividing the normalized pulse wave value by the sum of the discrete metric of the same identifier-aligned sampling set and a preset positive constant.

[0021] In this embodiment, the generation of the pulse phase sequence specifically includes: Local smoothing and waveform extremum extraction are performed on the pulse position hierarchical sampling sequence grouped by pulse position identifier and hierarchical identifier to obtain a set of periodic candidate labels; Local smoothing suppresses high-frequency spikes while maintaining the relative order of the main peak and trough positions of the pulse wave. Waveform extremum extraction compares the normalized pulse wave value sequence under the same pulse position and level label point by point, records the sampling points that meet the peak and trough conditions, and writes the peak position, trough position, corresponding unified time index, and corresponding pulse position and level label into the period candidate label set. The pulsation cycle is located by performing the correspondence between adjacent valley points and the main peak in the candidate periodic marker set, thus forming a set of single-beat cycles; The pulsation cycle localization uses the condition that there is a unique main peak between two adjacent valleys and the cycle duration is within the continuous fluctuation range of the same pulse position identifier and the same level identifier as the retention condition. Candidate cycles that do not meet the unique main peak condition, have abnormal cycle duration, or have reversed peak-valley order are eliminated. The retained cycle start position, cycle end position, main peak position, corresponding unified time index range, and corresponding pulse position identifier and level identifier are encapsulated into a single beat cycle set. For a single-wave cycle set, the starting inflection point, the rising end point, the peak end point, and the falling end point are extracted to obtain a set of phase boundary results. The starting and turning points are determined by the first stable upward starting point after the cycle's initial position; the ending point of the upward trend is determined by the last continuous upward sampling point before the main peak position; the ending point of the peak trend is determined by the boundary position where the waveform transitions from a plateau to a continuous decline after the main peak position; and the ending point of the downward trend is determined by the position where the waveform first enters a low-amplitude, slowly changing range after the main peak position. Phase segmentation is performed on each single-beat cycle based on the phase boundary result set to obtain the phase segmentation result set; The phase segmentation is divided into the starting phase segment according to the starting position of the cycle to the starting inflection point, the rising phase segment according to the starting inflection point to the rising phase ending point, the peak phase segment according to the rising phase ending point to the peak phase ending point, the falling phase segment according to the peak phase ending point to the falling phase ending point, and the residual phase segment according to the falling phase ending point to the end position of the cycle. The unified start and end time index of each segment, the sampling point sequence within the segment, and the corresponding single-beat cycle identifier are written into the phase segmentation result set. Perform phase order verification and abnormal phase removal on the phase segmentation result set to form a valid phase segment set; The phase sequence verification requires that the starting phase, rising phase, peak phase, falling phase, and residual phase be arranged sequentially and continuously according to a unified time index, and each phase must contain at least one sampling point. Abnormal phases are removed by deleting phase segmentation results that have inter-segment overlap, inter-segment reversal, main peak falling into non-peak phase, or residual phase missing. The phase segmentation results that pass the verification are retained as the set of valid phase segments. The effective phase segment set is sequentially encapsulated according to the single beat cycle identifier, pulse position identifier, level identifier and phase order to generate a pulse phase sequence. Each sequence unit in the pulse phase sequence contains the starting segment, rising segment, peak segment, falling segment, and residual segment under the corresponding single beat cycle identifier, as well as the unified time index of the start and end of each phase segment, the sampling point sequence within the segment, the corresponding pulse position identifier, and the corresponding level identifier. The pulse phase sequence is output in ascending order according to the unified time index.

[0022] In this embodiment, the generation of the pulse closed-loop path set specifically includes: The pulse phase sequence is grouped and organized according to the single-beat cycle identifier, pulse position identifier, and level identifier. The starting segment, rising segment, peak segment, falling segment, and residual segment within the same single-beat cycle are extracted to form a closed-loop input set. Each input unit in the closed-loop construction input set is written with a single beat cycle identifier, pulse position identifier, level identifier, unified start and end time index of five phase segments, and the intra-segment sampling point sequence corresponding to the five phase segments, maintaining consistency with the phase order in the pulse phase sequence. Connect the starting segment, rising segment, peak segment, falling segment and residual segment in sequence according to the phase order in the closed-loop construction input set to obtain the phase connection result set; The phase connection continuously splices the end sampling point and the start sampling point of two adjacent phase segments in the same input unit, retaining the original arrangement order within each phase segment, and writing the spliced ​​continuous sampling point sequence, phase segment boundary position, corresponding single beat cycle identifier, pulse position identifier and level identifier into the phase connection result set. Perform a head-and-tail closure construction on the phase connection result set to obtain a head-and-tail closure candidate set; The first and last closed construction uses the first and last sampling points in the continuous sampling point sequence in the phase connection result set as the closing endpoints. A closed back connection segment is added after the last sampling point, so that the closed back connection segment is connected to the amplitude position corresponding to the first sampling point along the continuous order of the unified time index. The closed back connection segment and the continuous sampling point sequence are written into the first and last closed candidate set. The closed back connection segment only completes the path closure of the single beat cycle and does not change the original order of the starting segment, rising segment, peak segment, falling segment and residual segment. Perform closure validity checks and sequence consistency checks on the candidate sets of closed paths at both ends to form a set of closed valid paths; The closure validity check requires that the starting point of the closed loop segment is consistent with the end of the continuous sampling point sequence, and the ending point is consistent with the beginning of the continuous sampling point sequence. After closure, there should be no path breakage, reverse jump, or repeated backtracking. The sequence consistency check requires that the boundary positions of the starting segment, rising segment, peak segment, falling segment, and residual segment remain consistent before and after closure. Candidates for closure at the beginning and end that do not meet the closure validity check or sequence consistency check are eliminated, and the paths that pass the check are retained as the set of valid closed paths. Perform path index writing and phase mapping encapsulation on the closed valid path set to obtain the closed-loop path index set; The path index writing associates each path in the set of closed valid paths with a single beat cycle identifier, pulse position identifier, and level identifier. The phase mapping encapsulation writes the start and end positions of the starting segment, rising segment, peak segment, falling segment, and residual segment in the closed path, the corresponding sampling point interval, and the corresponding unified time index range, forming a closed-loop path index set. The closed-loop path index set is sequentially encapsulated according to the pulse position identifier, level identifier, and single beat cycle identifier to generate the pulse closed-loop path set.

[0023] In this embodiment, the generation of the pulse structure spectrum set specifically includes: Extract the phase segment boundary position, closed connection segment, pulse position identifier, level identifier and single beat cycle identifier from the pulse closed loop path set, and construct a pulse constraint index table according to the pairing of adjacent phase segments within the same pulse position identifier, pairing of adjacent pulse positions in the same hand, pairing of corresponding pulse positions on the left and right, pairing of the same pulse position identifier across levels, and pairing of adjacent single beat cycles under the same level of the same pulse position identifier. The pairing of adjacent phase segments within the same pulse position identifier includes the pairing of the starting phase segment and the rising phase segment, the rising phase segment and the peak phase segment, the peak phase segment and the falling phase segment, and the falling phase segment and the remaining phase segment. The pairing of adjacent pulse positions on the same hand includes the pairing of the left hand cun and the left hand guan, the left hand guan and the left hand chi, the right hand cun and the right hand guan, and the right hand guan and the right hand chi. The pairing of corresponding pulse positions on the left and right hands includes the pairing of the left hand cun and the right hand cun, the left hand guan and the right hand guan, and the left hand chi and the right hand chi. The pairing of the same pulse position identifier across levels includes the pairing of floating and middle, middle and sinking, and floating and sinking. The pairing of adjacent single beat cycles under the same level of the same pulse position identifier includes the pairing of two single beat cycles that are adjacent in the same time index order. The phase segment boundary position, closed connection segment position, pulse position identifier, level identifier, and single beat cycle identifier corresponding to each pairing object are written into the pulse constrained index table. According to the pulse constraint index table, perform path segmentation and path alignment on the pulse closed-loop path set to obtain the quotient logarithmic signature input set. The path segmentation divides each pulse closed-loop path into the following segments: the starting segment path segment, the rising segment path segment, the peak segment path segment, the falling segment path segment, the residual segment path segment, and the closed loop segment. The path alignment is performed according to the pairing objects already written in the pulse constraint index table. Only path segments with the same pairing type and consistent phase order are paired and encapsulated. The paired path segments, corresponding pairing types, corresponding pulse position identifiers, corresponding level identifiers, and corresponding single beat cycle identifiers are written into the quotient logarithmic signature input set. Perform equivalent merging and repetition elimination on the input set of quotient logarithmic signatures to obtain the intermediate set of quotient logarithmic signatures; The equivalent merging merges the path block pairing results with the same pairing type, the same pulse position identifier correspondence, the same level identifier correspondence, the same single beat cycle identifier correspondence, and the same phase order into the same merging unit. The duplicate path block pairing records after merging are deleted, and the unique path block pairing result after merging is retained as the intermediate set of quotient logarithm signature. Perform pulse-constrained quotient logarithmic signature transformation on the intermediate set of quotient logarithmic signatures to obtain the set of intra-pulse phase exchange terms, the set of inter-pulse exchange terms, the set of hierarchical migration terms, and the set of cross-cycle inheritance terms. The set of phase exchange terms within pulse positions, the set of exchange terms between pulse positions, the set of hierarchical migration terms, and the set of cross-cycle inheritance terms are written with term-level identifiers to obtain the set of pulse structure spectrum entries. The item-level identifier is written by writing the item type, corresponding pairing type, corresponding pulse position identifier, corresponding level identifier, corresponding single beat cycle identifier, corresponding path block source, and corresponding closed-loop path index position for each item. The pulse structure spectrum entries are uniformly encapsulated according to item type, pulse position identifier, level identifier, and single beat cycle identifier to generate a pulse structure spectrum set.

[0024] In this embodiment, the pulse constrained quotient logarithmic signature transformation reorganizes the path block pairing results corresponding to each merging unit in the intermediate set of quotient logarithmic signatures according to a unified time index order, and obtains the local path of the merging unit. The ordered integral combination within a preset order is calculated for the local path of the merging unit, and a logarithmic mapping is applied to convert it into a local logarithmic signature representation within a preset order. The corresponding coordinates in the local logarithmic signature representation are selectively read according to the pairing type corresponding to the merging unit. Among them, reading the pairing of adjacent phase segments within the same pulse position identifier obtains the phase exchange coordinates within the pulse position; reading the pairing of adjacent pulse positions in the same hand and the pairing of corresponding pulse positions on the left and right sides obtains the exchange coordinates between pulse positions; reading the pairing of cross-levels within the same pulse position identifier obtains the level migration coordinates; and reading the pairing of adjacent single beat cycles under the same level within the same pulse position identifier obtains the cross-cycle inheritance coordinates.

[0025] In this embodiment, the generation of the closed-loop residual results specifically includes: Extract the closed loop segments, first sampling point, last sampling point, phase segment boundary position, pulse position identifier, hierarchical identifier and single beat cycle identifier from the pulse closed loop path set, and extract the intra-pulse phase exchange term, inter-pulse exchange term, hierarchical migration term and cross-cycle inheritance term from the pulse structure spectrum set to form the residual calculation input set; Perform head-and-tail closure deviation calculation and phase boundary deviation calculation on the deviation calculation input set to obtain the single-shot closed-loop deviation; The calculation of the first and last closure deviation reads the amplitude difference between the end point and the first sampling point of the closed loop segment in the same pulse closed loop path, the amplitude difference between the start point and the last sampling point of the closed loop segment, and the continuous change difference of the closed loop segment itself. The calculation of the phase boundary deviation reads the boundary position difference and boundary sequence difference of the starting segment, rising segment, peak segment, falling segment and residual segment before and after closure. After being merged according to the same single beat cycle, the single beat closed loop deviation is formed. Extract inter-pulse position exchange terms and hierarchical migration terms from the pulse structure spectrum set, perform grouping and aggregation according to pairing type, and obtain the exchange sub-spectrum input set; The grouping and aggregation process includes classifying the exchange items between the pulse positions corresponding to the left hand cun and left hand guan, left hand guan and left hand chi, right hand cun and right hand guan, and right hand guan and right hand chi into the adjacent pulse position group of the same hand; classifying the exchange items between the pulse positions corresponding to the left hand cun and right hand cun, left hand guan and right hand guan, and left hand chi and right hand chi into the corresponding pulse position group of the left and right hands; and classifying the hierarchical migration items corresponding to the superficial and middle selection, middle selection and deep selection, and superficial selection and deep selection under the same pulse position identifier into the cross-level pulse position group. Perform directional pairing, reverse cancellation, and spectral encapsulation on the input set of exchanged sub-spectrums to obtain the pulse position exchanged sub-spectrums; The direction pairing is matched one-to-one with the exchange items in the same group that have opposite exchange directions and the corresponding pulse position identifier, level identifier, and single beat cycle identifier are consistent. The reverse cancellation uses the intensity difference of the exchange items with opposite directions as the exchange sub-intensity. The positive or negative sign of the exchange item intensity indicates the exchange direction. The order of the exchange sub-intensities in the same group forms a spectral sequence. The exchange sub-spectrum of adjacent pulse positions in the same hand, the exchange sub-spectrum of corresponding pulse positions on the left and right, and the exchange sub-spectrum of pulse positions across levels are output respectively, and are uniformly packaged into a pulse position exchange sub-spectrum. The single-beat closed-loop deviation, pulse position exchange sub-spectrum, pulse position phase exchange term and cross-cycle inheritance term are jointly merged according to the same pulse position identifier, the same level identifier and the same single-beat cycle identifier to generate path deviation decomposition results; The path deviation decomposition results include the first and last closure deviation components, phase boundary deviation components, phase exchange residual components within the pulse position, exchange residual components of adjacent pulse positions in the same hand, exchange residual components of corresponding pulse positions on the left and right, cross-level pulse position exchange residual components, cross-cycle inherited residual components, and dominant residual source items. Among them, the first and last closure deviation components and the phase boundary deviation components are obtained by writing the first and last closure deviation calculation results and the phase boundary deviation calculation results in the single-beat closed-loop deviation amount, respectively. The pulse position phase exchange residual components are obtained by merging the pulse position phase exchange items corresponding to adjacent phase segments within the same pulse position identifier according to the phase order. The same hand adjacent pulse position exchange residual components, left and right corresponding pulse position exchange residual components and cross-level pulse position exchange residual components are obtained by merging the same hand adjacent pulse position exchange sub-spectrum, left and right corresponding pulse position exchange sub-spectrum and cross-level pulse position exchange sub-spectrum according to the pulse position identifier and level identifier, respectively. The cross-cycle inheritance residual components are obtained by merging the cross-cycle inheritance items corresponding to adjacent single-beat cycles under the same pulse position identifier and level according to the single-beat cycle order. The dominant residual source item is determined by the residual source corresponding to the component with the largest absolute value among all components. Based on the path deviation decomposition results, residual adjudication and residual classification are performed to generate closed-loop residual results.

[0026] In this embodiment, the residual determination is based on whether the single-beat closed-loop deviation exceeds a preset closure deviation threshold, whether the phase exchange residual component within the pulse position exceeds a preset phase exchange residual threshold, whether the exchange residual components of adjacent pulse positions on the same hand and the exchange residual components of corresponding pulse positions on the left and right exceed a preset pulse position exchange residual threshold, whether the cross-level pulse position exchange residual component exceeds a preset level migration residual threshold, and whether the cross-cycle inheritance residual component exceeds a preset inheritance residual threshold. Specifically, when the single-beat closed-loop deviation does not exceed the preset closure deviation threshold and each exchange residual component does not exceed the corresponding threshold, the decision is a complete closed-loop state. When the single-beat closed-loop deviation exceeds the preset closure deviation threshold and each exchange residual component does not exceed the corresponding threshold, the decision is a tense closed-loop state. When the single-beat closed-loop deviation does not exceed the preset closure deviation threshold and at least one exchange residual component exceeds the corresponding threshold, the decision is a deviated closed-loop state. When the single-beat closed-loop deviation exceeds the preset closure deviation threshold and at least two exchange residual components exceed the corresponding threshold, the decision is a broken closed-loop state. The closed-loop state obtained from the decision, the dominant residual source item, the corresponding pulse position identifier, the corresponding level identifier, and the corresponding single-beat cycle identifier are all written into the closed-loop residual result.

[0027] In this embodiment, the generation of pulse type results and structural interpretation results specifically includes: Extract the intra-pulse phase exchange term, inter-pulse exchange term, hierarchical transfer term, and cross-cycle inheritance term from the pulse structure spectrum set. Extract the same-hand adjacent pulse exchange sub-spectrum, left-right corresponding pulse exchange sub-spectrum, and cross-hierarchical pulse exchange sub-spectrum from the pulse exchange sub-spectrum. Extract the closed-loop state and dominant residual source term from the closed-loop residual result to form the recognition input set. Based on the input set, perform category prototype matching to obtain a candidate set of pulse category; The category prototype matching generates corresponding pulse category candidate entries for each recognition input unit according to the preset pulse category prototype, and encapsulates them into a pulse category candidate set. The preset pulse category prototype is a template set consisting of pulse category name and corresponding pulse position phase exchange item combination, pulse position exchange item combination, hierarchical migration item combination, cross-cycle inheritance item combination, pulse position exchange sub-spectrum combination and closed loop state. Perform category strength aggregation and category sorting on the candidate set of pulse categories to obtain a set of category score results; The category strength aggregation method accumulates the hit counts of all candidate entries for the same pulse category name to obtain the category strength value for the corresponding pulse category name. The categories are sorted from high to low according to the category strength value and written into the category order to obtain the category score result set. Based on the identification input set and the category scoring result set, dominant exchange sub-items, pulse position imbalance items, hierarchical transfer items and cross-period inheritance items are extracted to generate a candidate set of structural interpretations; The dominant exchange term is jointly determined by the pulse position exchange sub-spectrum and the dominant residual source term in the identification input set. The pulse position imbalance term is determined by pairing the pulse positions corresponding to the largest absolute value of the exchange sub-intensity in the pulse position exchange sub-spectrum of adjacent pulse positions in the same hand and the pulse position exchange sub-spectrum of the left and right corresponding pulse positions. The hierarchical migration term is determined by the entry in the hierarchical migration term that corresponds to the cross-hierarchical pulse position exchange residual component and has the largest absolute value. The cross-cycle inheritance term is determined by the entry in the cross-cycle inheritance term that corresponds to the cross-cycle inheritance residual component and has the largest absolute value. These terms, along with the corresponding pulse position identifier, hierarchical identifier, and single beat cycle identifier, are encapsulated into a structural interpretation candidate set. A consistency check and result adjudication are performed on the category scoring result set and the structural interpretation candidate set to obtain the pulse recognition result set. The consistency check reads the pulse category candidate entries with the highest category order in the category scoring result set, and reads the dominant exchange sub-items, pulse position imbalance items, hierarchical migration items, and cross-cycle inheritance items in the structural interpretation candidate set. It checks whether the correspondence between the pulse category candidate entries and the structural interpretation candidate set is consistent in terms of pulse position identifier, hierarchical identifier, single beat cycle identifier, and dominant residual source item. Consistent candidate results are directly retained, and inconsistent candidate results are re-determined according to category order and dominant residual source item to generate a pulse recognition result set. Output the pulse category result and the structural interpretation result according to the adjudication result in the pulse recognition result set; The pulse category result is obtained by writing the highest-ranking entry of the category retained after adjudication in the pulse recognition result set. The structural interpretation result is obtained by writing the dominant exchange sub-item, pulse position imbalance item, hierarchical migration item and cross-cycle inheritance item corresponding to the entry.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a joint diagnosis and treatment scenario involving the preventive medicine clinic and the internal medicine department of a traditional Chinese medicine hospital. In this scenario, patients included those with clearly defined chronic discomfort, as well as those who wished to undergo constitution identification and pulse tracking after a physical examination. A common problem in previous outpatient clinics was that while digital pulse diagnosis equipment could collect pulse waves, subsequent analysis often remained at a superficial level, focusing on indicators such as amplitude, rhythm, and peak / trough positions. When faced with differences in the linkage between the left and right wrist cun, guan, and chi positions of the same patient, variations in the levels of palpation (superficial, middle, and deep), and the continuity of continuous pulsations, only a rough judgment could be given. It was difficult to explain why complex pulse characteristics such as wiry, slippery, thin, and hesitant pulses appeared, and it was also difficult to transform the positional, force, and level changes in the physician's experience into a traceable data chain. To address this issue, during outpatient visits, pulse wave data is continuously collected from the cun, guan, and chi positions of both hands under superficial, middle, and deep palpation techniques. Time alignment, hierarchical segmentation, anomaly removal, and amplitude normalization are performed to form a stratified pulse position sampling sequence. Subsequently, the continuous pulsations at each pulse position and level are located, and the initial, rising, peak, falling, and residual phases are defined to form a pulse phase sequence. The single-beat cycle is then closed at both ends to form a pulse closed-loop path set. Based on this, a logarithmic signature transformation of the pulse constraint quotient is performed on the closed-loop path to extract intra-pulse phase exchange terms, inter-pulse exchange terms, hierarchical migration terms, and cross-cycle inheritance terms, forming a pulse structure spectrum set. The single-beat closed-loop deviation, pulse position exchange sub-spectrum, and closed-loop residual results are then calculated, ultimately outputting pulse category results and structural interpretation results.

[0029] In practical applications, outpatient clinics simultaneously retain manual pulse diagnosis records, symptom descriptions from electronic medical records, tongue image collection records, and repeated data collection from the same patient at different times of visit. This data is used to cross-verify the stability and interpretability of the invention's output. Data comparison shows that when the patient's pulse characteristics are relatively simple, the invention can stably output a consistent pulse category and provide a dominant exchange sub-item corresponding to the physician's palpation sensation. When the patient exhibits significant differences between the left and right hands, inconsistent changes in the superficial, middle, and deep pulse levels, or unstable fluctuations in the pulse during the same collection, the invention does not simply provide a single category. Instead, it breaks down the source of deviation through pulse position imbalance items, level migration items, and cross-cycle inheritance items, allowing physicians to intuitively see whether the abnormality mainly stems from pulse position coordination imbalance, abnormal level migration, or unstable cycle continuity. Continuous follow-up data also shows that when symptom descriptions change slowly but subtle fluctuations occur in the internal structure of the pulse, traditional pulse wave parameter methods often fail to identify these fluctuations. This invention, however, can first reflect the deviations in the initial and final closure and phase boundary in the closed-loop residual results, and then indicate the sources of change in the relevant pulse position and level in the structural interpretation results, providing a basis for subsequent diagnostic adjustments. This demonstrates that this invention can not only intelligently identify TCM pulse diagnosis data, but also express the internal structural relationships of complex pulses in a clear data form, effectively solving the problems of insufficient utilization of structural information, weak result interpretability, and insufficient stability in identifying complex pulses in existing digital pulse diagnosis methods.

[0030] Table 1. Comparison of Comprehensive Performance of Pulse Structured Intelligent Analysis

[0031] As shown in Table 1, the method of this invention significantly outperforms the traditional pulse wave parameter method and the conventional time-series classification method in three core indicators: accuracy of pulse type identification, accuracy of composite pulse identification, and consistency rate of repeated sampling from the same subject. In particular, the accuracy of composite pulse identification reaches 88.16%, which is 19.89 percentage points higher than the traditional pulse wave parameter method and 10.21 percentage points higher than the conventional time-series classification method. This indicates that relying solely on superficial features such as amplitude, peaks and troughs, or rhythm can easily lead to excessive information compression when dealing with complex pulse combinations such as those combining wiry and slippery pulses or thin and rough pulses. In contrast, this invention, through refined modeling of the initial, rising, peak, falling, and residual phases, completely preserves the sequential changes within a single pulse cycle, thus more fully reflecting the internal structure of complex pulses. The accuracy rate of pulse type identification reached 92.48%, and the consistency rate of repeated collections of the same subject reached 93.21%, indicating that the present invention has good result stability under multiple collection conditions. This is directly related to the combined effect of the construction of the pulse closed-loop path set, the extraction of quotient logarithmic signature constraints, and the verification of closed-loop residuals.

[0032] Further examination of the accuracy rates for identifying differences in pulse positions between the left and right hands, the accuracy rates for identifying changes in the superficial, middle, and deep pulse levels, and the accuracy rates for phase division within a single pulse cycle reveals that the method of this invention achieves accuracy rates of 90.44%, 89.73%, and 91.35%, respectively, all at a high level. This result indicates that this invention can not only identify pulse types but also effectively express the synergistic relationships between pulse positions and the migration relationships between levels. Traditional pulse wave parameter methods often compress multi-pulse information into several statistical quantities, easily weakening the linkage differences between the cun, guan, and chi positions. While conventional temporal classification methods outperform traditional methods in overall recognition rate, they lack explicit phase division and closed-loop structure expression, making them insufficiently sensitive to the transitional relationships between superficial, middle, and deep pulses and subtle imbalances between corresponding pulse positions on the left and right hands. This invention characterizes the internal sequence of a single beat, the coupling between pulse positions, and the hierarchical changes through intra-pulse phase exchange terms, inter-pulse exchange terms, and hierarchical migration terms, respectively. This allows the differences between the left and right hands and the variations in superficiality, mid-depth, and depth to no longer remain at the level of implicit features, but to be transformed into calculable, comparable, and traceable structural results. Therefore, the improvement is more significant in these indicators that embody the core characteristics of TCM pulse diagnosis.

[0033] In terms of the consistency rate between structural interpretation and physician consensus, the consistency rate of abnormality source localization, and the abnormality detection rate of closed-loop residuals, the method of this invention achieved 87.52%, 85.34%, and 90.27%, respectively, showing a more significant improvement. This indicates that the advantage of this invention lies not only in improving the accuracy of identification, but also in its ability to map the identification conclusions to specific structural sources. Although the traditional pulse wave parameter method can provide several waveform parameters, it is difficult to explain whether the abnormality originates from insufficient single-beat closure, pulse position coordination imbalance, hierarchical migration abnormality, or cross-cycle inheritance instability. Although the conventional temporal classification method can output category judgments, it usually lacks clear explanations for the position, potential, hierarchy, and continuous evolution relationships that are of particular concern in TCM clinical practice. This invention, through pulse position exchange sub-spectrum and closed-loop residual results, decomposes the first and last closure deviation, phase boundary deviation, intra-pulse position phase exchange residual, inter-pulse position exchange residual, cross-hierarchical migration residual, and cross-cycle inheritance residual, and further outputs the dominant exchange sub-item, pulse position imbalance item, hierarchical migration item, and cross-cycle inheritance item, allowing physicians to directly see which type of structural relationship the abnormality mainly falls on. This output method, which extends from category results to structural interpretation results, makes the present invention more usable in clinical applications and better meets the actual needs of traditional Chinese medicine pulse diagnosis, which emphasizes comprehensive judgment and syndrome differentiation.

[0034] From the perspective of average analysis time per case, the method of this invention takes 1.63 seconds, which is slightly higher than the conventional time-series classification method of 1.47 seconds, but still significantly faster than the traditional pulse wave parameter method of 1.92 seconds. This indicates that after introducing multi-layered structural calculations such as pulse closed-loop path, pulse constraint quotient logarithmic signature transformation, pulse position exchange sub-spectrum, and closed-loop residual results, the overall processing efficiency remains at a high level, without excessive time overhead due to enhanced structural representation. This result demonstrates that while ensuring recognition accuracy, stability, and interpretability, this invention also possesses good engineering feasibility and can meet the real-time requirements of outpatient auxiliary analysis and continuous follow-up assessment.

[0035] As can be seen from Table 1, this invention transforms multi-pulse position and multi-level pulse wave data into pulse phase sequences and pulse closed-loop path sets. Then, it uses the logarithmic signature transformation of pulse constraint quotients to extract phase exchange terms within pulse positions, exchange terms between pulse positions, hierarchical transfer terms, and cross-cycle inheritance terms. Combined with the pulse position exchange sub-spectrum and closed-loop residual results, it completes pulse identification and structural interpretation. This not only effectively improves the accuracy of pulse category identification and stability in complex pulse scenarios, but also significantly enhances the ability to express differences between left and right hand pulse positions, changes in the superficial, middle, and deep levels, the internal sequence relationship of a single beat cycle, and the source of abnormalities. It shows superior comprehensive effects in terms of identification accuracy, result consistency, abnormality localization ability, and clinical interpretability.

[0036] The above are merely preferred embodiments 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 analysis of TCM pulse diagnosis data based on pattern recognition, characterized in that, Includes the following steps: Acquire pulse wave data of the left and right hands at the Cun, Guan, and Chi positions under the floating, middle, and deep positions, align them according to the pulse position identifier, level identifier, and time index, and generate a pulse position hierarchical sampling sequence. Perform pulsation cycle localization and phase division on the pulse position layered sampling sequence to generate a pulse phase sequence containing the onset segment, rising segment, peak segment, falling segment and residual segment; Based on the pulse phase sequence, the single beat cycle of each pulse position and level is closed at the beginning and end while maintaining the phase order, generating a set of closed-loop pulse paths. Perform a pulse constraint quotient logarithmic signature transformation on the pulse closed-loop path set to extract intra-pulse phase exchange terms, inter-pulse exchange terms, hierarchical migration terms, and cross-period inheritance terms to generate a pulse structure spectrum set. The single-beat closed-loop deviation is calculated based on the pulse closed-loop path set, and the pulse exchange sub-spectrum of adjacent pulse positions in the same hand, corresponding pulse positions on the left and right and cross-level pulse positions is calculated based on the pulse structure spectrum set. The closed-loop residual result is generated by combining the single-beat closed-loop deviation. Pulse image recognition is performed based on the pulse image structure spectrum set, pulse position exchange sub-spectrum, and closed-loop residual results. The output pulse image category results and structure interpretation results are provided. The structure interpretation results include the corresponding dominant exchange sub-items, pulse position imbalance items, hierarchical migration items, and cross-cycle inheritance items.

2. The intelligent analysis method for TCM pulse diagnosis data based on pattern recognition according to claim 1, characterized in that, The generation of the pulse position hierarchical sampling sequence specifically includes: Pulse wave data, sampling timestamps, and pulse pressure values ​​were collected at the left hand cun, left hand guan, left hand chi, right hand cun, right hand guan, and right hand chi positions respectively, under the conditions of superficial, middle, and deep sampling, to form the original sampling set for pulse diagnosis; Perform pulse diagnosis hierarchical segmentation and stable segment extraction on the original pulse diagnosis sampling set to obtain a set of hierarchical effective sampling segments; Anomaly removal and baseline trimming are performed on the set of effective sampled segments at each level to obtain a set of purified pulse wave segments; Perform uniform sampling interval resampling and time index reconstruction on the purified pulse wave fragment set to obtain a time-aligned sample set; Based on the hand-side position, pulse position, pulse retrieval level, and unified time index, the time-aligned sampling set is subjected to identifier binding and sequential encapsulation to obtain the identifier-aligned sampling set; Amplitude normalization is performed on the identifier-aligned sampling set, and the results are output according to the pulse position identifier, level identifier, and unified time index to generate a pulse position hierarchical sampling sequence.

3. The intelligent analysis method for TCM pulse diagnosis data based on pattern recognition according to claim 1, characterized in that, The generation of the pulse phase sequence specifically includes: Local smoothing and waveform extremum extraction are performed on the pulse position hierarchical sampling sequence grouped by pulse position identifier and hierarchical identifier to obtain a set of periodic candidate labels; The pulsation cycle is located by performing the correspondence between adjacent valley points and the main peak in the candidate periodic marker set, thus forming a set of single-beat cycles; For a single-wave cycle set, the starting inflection point, the rising end point, the peak end point, and the falling end point are extracted to obtain a set of phase boundary results. Phase segmentation is performed on each single-beat cycle based on the phase boundary result set to obtain the phase segmentation result set; Perform phase order verification and abnormal phase removal on the phase segmentation result set to form a valid phase segment set; The effective phase segment set is sequentially encapsulated according to the single beat cycle identifier, pulse position identifier, level identifier and phase order to generate a pulse phase sequence.

4. The intelligent analysis method for TCM pulse diagnosis data based on pattern recognition according to claim 1, characterized in that, The generation of the pulse closed-loop path set specifically includes: The pulse phase sequence is grouped and organized according to the single-beat cycle identifier, pulse position identifier, and level identifier. The starting segment, rising segment, peak segment, falling segment, and residual segment within the same single-beat cycle are extracted to form a closed-loop input set. Connect the starting segment, rising segment, peak segment, falling segment and residual segment in sequence according to the phase order in the closed-loop construction input set to obtain the phase connection result set; Perform a head-and-tail closure construction on the phase connection result set to obtain a head-and-tail closure candidate set; Perform closure validity checks and sequence consistency checks on the candidate sets of closed paths at both ends to form a set of closed valid paths; Perform path index writing and phase mapping encapsulation on the closed valid path set to obtain the closed-loop path index set; The closed-loop path index set is sequentially encapsulated according to the pulse position identifier, level identifier, and single beat cycle identifier to generate the pulse closed-loop path set.

5. The method for intelligent analysis of TCM pulse diagnosis data based on pattern recognition according to claim 1, characterized in that, The generation of the pulse structure spectrum set specifically includes: Extract the phase segment boundary position, closed connection segment, pulse position identifier, level identifier and single beat cycle identifier from the pulse closed loop path set, and construct a pulse constraint index table according to the pairing of adjacent phase segments within the same pulse position identifier, pairing of adjacent pulse positions in the same hand, pairing of corresponding pulse positions on the left and right, pairing of the same pulse position identifier across levels, and pairing of adjacent single beat cycles under the same level of the same pulse position identifier. According to the pulse constraint index table, perform path segmentation and path alignment on the pulse closed-loop path set to obtain the quotient logarithmic signature input set. Perform equivalent merging and repetition elimination on the input set of quotient logarithmic signatures to obtain the intermediate set of quotient logarithmic signatures; Perform pulse-constrained quotient logarithmic signature transformation on the intermediate set of quotient logarithmic signatures to obtain the set of intra-pulse phase exchange terms, the set of inter-pulse exchange terms, the set of hierarchical migration terms, and the set of cross-cycle inheritance terms. The set of phase exchange terms within pulse positions, the set of exchange terms between pulse positions, the set of hierarchical migration terms, and the set of cross-cycle inheritance terms are written with term-level identifiers to obtain the set of pulse structure spectrum entries. The pulse structure spectrum entries are uniformly encapsulated according to item type, pulse position identifier, level identifier, and single beat cycle identifier to generate a pulse structure spectrum set.

6. The method for intelligent analysis of TCM pulse diagnosis data based on pattern recognition according to claim 5, characterized in that, The pulse constrained quotient logarithmic signature transformation reorganizes the path block pairing results corresponding to each merging unit in the intermediate set of quotient logarithmic signatures according to a unified time index order, obtaining the local path of the merging unit. The ordered integral combination within a preset order is calculated for the local path of the merging unit, and a logarithmic mapping is applied to convert it into a local logarithmic signature representation within a preset order. The corresponding coordinates in the local logarithmic signature representation are selectively read according to the pairing type corresponding to the merging unit. Among them, reading the pairing of adjacent phase segments within the same pulse position identifier obtains the phase exchange coordinates within the pulse position; reading the pairing of adjacent pulse positions in the same hand and the pairing of corresponding pulse positions on the left and right sides obtains the exchange coordinates between pulse positions; reading the pairing of cross-levels within the same pulse position identifier obtains the level migration coordinates; reading the pairing of adjacent single beat cycles under the same level within the same pulse position identifier obtains the cross-cycle inheritance coordinates.

7. The intelligent analysis method for TCM pulse diagnosis data based on pattern recognition according to claim 1, characterized in that, The generation of the closed-loop residual results specifically includes: Extract the closed loop segments, first sampling point, last sampling point, phase segment boundary position, pulse position identifier, hierarchical identifier and single beat cycle identifier from the pulse closed loop path set, and extract the intra-pulse phase exchange term, inter-pulse exchange term, hierarchical migration term and cross-cycle inheritance term from the pulse structure spectrum set to form the residual calculation input set; Perform head-and-tail closure deviation calculation and phase boundary deviation calculation on the deviation calculation input set to obtain the single-shot closed-loop deviation; Extract inter-pulse position exchange terms and hierarchical migration terms from the pulse structure spectrum set, perform grouping and aggregation according to pairing type, and obtain the exchange sub-spectrum input set; Perform directional pairing, reverse cancellation, and spectral encapsulation on the input set of exchanged sub-spectrums to obtain the pulse position exchanged sub-spectrums; The single-beat closed-loop deviation, pulse position exchange sub-spectrum, pulse position phase exchange term and cross-cycle inheritance term are jointly merged according to the same pulse position identifier, the same level identifier and the same single-beat cycle identifier to generate path deviation decomposition results; Based on the path deviation decomposition results, residual adjudication and residual classification are performed to generate closed-loop residual results.

8. The intelligent analysis method for TCM pulse diagnosis data based on pattern recognition according to claim 7, characterized in that, The residual determination is based on whether the single-beat closed-loop deviation exceeds a preset closure deviation threshold, whether the phase exchange residual component within the pulse position exceeds a preset phase exchange residual threshold, whether the exchange residual components of adjacent pulse positions on the same hand and the exchange residual components of corresponding pulse positions on the left and right exceed preset pulse position exchange residual thresholds, whether the cross-level pulse position exchange residual component exceeds a preset level migration residual threshold, and whether the cross-cycle inheritance residual component exceeds a preset inheritance residual threshold. Specifically, when the single-beat closed-loop deviation does not exceed the preset closure deviation threshold and each exchange residual component does not exceed the corresponding threshold, the decision is a complete closed-loop state. When the single-beat closed-loop deviation exceeds the preset closure deviation threshold and each exchange residual component does not exceed the corresponding threshold, the decision is a tense closed-loop state. When the single-beat closed-loop deviation does not exceed the preset closure deviation threshold and at least one exchange residual component exceeds the corresponding threshold, the decision is a deviated closed-loop state. When the single-beat closed-loop deviation exceeds the preset closure deviation threshold and at least two exchange residual components exceed the corresponding threshold, the decision is a broken closed-loop state. The closed-loop state obtained from the decision, the dominant residual source item, the corresponding pulse position identifier, the corresponding level identifier, and the corresponding single-beat cycle identifier are all written into the closed-loop residual result.

9. The intelligent analysis method for TCM pulse diagnosis data based on pattern recognition according to claim 1, characterized in that, The generation of the pulse category results and structural interpretation results specifically includes: Extract the intra-pulse phase exchange term, inter-pulse exchange term, hierarchical transfer term, and cross-cycle inheritance term from the pulse structure spectrum set. Extract the same-hand adjacent pulse exchange sub-spectrum, left-right corresponding pulse exchange sub-spectrum, and cross-hierarchical pulse exchange sub-spectrum from the pulse exchange sub-spectrum. Extract the closed-loop state and dominant residual source term from the closed-loop residual result to form the recognition input set. Based on the input set, perform category prototype matching to obtain a candidate set of pulse category; Perform category strength aggregation and category sorting on the candidate set of pulse categories to obtain a set of category score results; Based on the identification input set and the category scoring result set, dominant exchange sub-items, pulse position imbalance items, hierarchical transfer items and cross-period inheritance items are extracted to generate a candidate set of structural interpretations; A consistency check and result adjudication are performed on the category scoring result set and the structural interpretation candidate set to obtain the pulse recognition result set. Output the pulse category result and the structural interpretation result according to the adjudication results in the pulse recognition result set.

Images