Fiber Optic Driven Neural Network Pattern Recognition Method and System

Abstract

This invention relates to the field of pattern recognition technology, specifically to a fiber-optic driven neural network pattern recognition method and system, comprising the following steps: acquiring an optical interference signal, dividing it into periodic groups to extract amplitude and phase trajectories, constructing a cooperative trajectory group and identifying synchronous evolution patterns, generating a structure mapping map to extract trajectory codes for input into the neural network recognition layer, detecting interference pattern perturbations to remove unstable paths, and outputting recognition results with matching optical power. In this invention, the periodic structural characteristics of the optical interference signal are introduced to divide the input data into path trajectories. A cooperative evolution structure is established using the multidimensional variation trends of amplitude and phase, enhancing the ability to characterize the intrinsic relationships between dynamic signal features. Combined with a neural network retrieval mechanism, the recognition accuracy of evolution patterns is effectively improved. By analyzing the matching relationship between probability output and optical power fluctuations in the recognition path, the recognition output results are ensured to have stronger response consistency and classification effectiveness.

Description

Technical Field

[0001] This invention relates to the field of pattern recognition technology, and in particular to a fiber-optic driven neural network pattern recognition method and system. Background Technology

[0002] Pattern recognition technology involves using computational methods to analyze input signals or data and automatically identify the categories, structures, or patterns they represent. Core aspects include feature extraction methods, classification decision-making methods, training and learning mechanisms, input signal preprocessing methods, and the optimized design of recognition algorithms and strategies. This field is widely used in various information processing areas such as image recognition, speech recognition, biometric recognition, and text recognition. To date, it has developed into a system of various recognition methods based on machine learning, particularly neural networks. Especially within the deep learning framework, using multi-layered neural networks to perform hierarchical abstract learning of input data has become the mainstream approach. This is accompanied by a trend towards diversification of perception methods, including replacing or supplementing traditional electrical signal input methods to adapt to the recognition needs of different application scenarios. Traditional neural network pattern recognition methods refer to recognition methods that use artificial neural network structures composed of multi-layered neurons to process and classify input data. This mainly involves receiving signal data through the input layer, performing nonlinear mapping through the hidden layer, and forming the recognition result at the output layer. The process requires iterative optimization of weights and biases using batches of sample data during the training phase to enable the network to learn features and classification rules from the data. Traditional methods employ digital signal preprocessing techniques for noise suppression and feature extraction, then input the processed data into a neural network model for recognition and judgment. In terms of implementation, these methods rely on electronic communication to transmit information, and their recognition performance is highly dependent on the quality of signal acquisition and the completeness of the model training data.

[0003] Traditional neural network pattern recognition methods rely on electrical signals as the primary input. During signal acquisition, these methods are susceptible to electromagnetic interference, background noise, and sensor sensitivity limitations, leading to unstable input signal quality and consequently affecting recognition accuracy. While feature extraction is performed through preprocessing, it is difficult to accurately capture the temporal evolution patterns and subtle perturbation trends in the input signal. Although the hidden layer structure possesses certain nonlinear mapping capabilities, it lacks a targeted modeling mechanism for the collaborative relationships between multidimensional dynamic features. The training process is highly dependent on the integrity of the samples, and recognition performance fluctuates in dynamic scenarios. It cannot effectively adapt to the stable recognition requirements in multi-source sensing and high-noise scenarios, and overall exhibits significant limitations in handling continuous dynamic changes and signal perturbations. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a fiber-optic driven neural network pattern recognition method and system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a fiber-optic driven neural network pattern recognition method, comprising the following steps:

[0006] S1: Based on the optical interference signal data stream collected by the optical fiber sensing node, it is divided into continuous periodic groups according to the periodic oscillation characteristics of the light wave. The extreme value center trajectory in the amplitude change within each group is extracted and the time sequence path is recorded to generate an optical interference cooperative trajectory group.

[0007] S2: Call the amplitude path and phase offset path in the optical interference cooperative trajectory group, detect whether the change direction of the two types of trajectories between adjacent period groups maintains a synchronous trend, mark the trajectory cluster with continuous unidirectional evolution behavior as cooperative evolution mode, and generate a trajectory cooperative structure mapping map.

[0008] S3: Based on the trajectory cluster features marked in the trajectory cooperative structure mapping map, extract the morphological code of the path trajectory and input it into the neural network recognition layer, retrieve path nodes with the same trajectory pattern label in the neural network sample set, and generate a neural network recognition channel set.

[0009] S4: Call the neural network to identify the channel set, collect the state change information of the edge pixels of the light stripes in the interference pattern area in multiple frames of images, detect the trend of disturbance feature changes, and generate a structure map after path interference is eliminated.

[0010] As a further embodiment of the present invention, the optical interference cooperative trajectory group includes trajectory node association, trajectory periodic continuity information, and path synchronization offset features; the trajectory cooperative structure mapping diagram includes path clustering structure, cooperative label nodes, and mapping evolution trend features; the neural network recognition channel set includes encoded path set, neural response state, and target pattern channel index; and the structure diagram after path interference elimination includes stable recognition path set, image perturbation shielding information, and channel label set.

[0011] As a further aspect of the present invention, the step of obtaining the optical interference cooperative trajectory set specifically includes:

[0012] S111: Based on the optical interference signal data stream collected by the optical fiber sensing node, detect the boundary features of the oscillation period of the signal, extract the periodically adjacent extreme point pairs, calculate the difference between the mean and peak values ​​of the oscillation within the continuous period, classify them to form period intervals, record the period interval index and oscillation amplitude characteristic parameters of each group, and generate a periodic oscillation amplitude index group.

[0013] S112: Call the periodic oscillation amplitude index group, filter the extreme center trajectory within the periodic group, extract continuous extreme points in amplitude changes, calculate the amplitude offset trajectory transformation rate, and obtain the offset trajectory change trend.

[0014] S113: Based on the trend of the offset trajectory change, identify the phase angle point corresponding to the peak transfer point, record the position sequence and angle change sequence in the continuous period group, construct the corresponding trajectory segment direction vector group, and combine it with the original amplitude center trajectory sequence to generate the optical interference cooperative trajectory group.

[0015] As a further aspect of the present invention, the amplitude offset trajectory transformation rate is expressed by the formula:

[0016] ;

[0017] in, Represents the amplitude offset trajectory change rate. Representing the Extreme amplitude values ​​within a periodic segment Representing the Extreme amplitude values ​​within a periodic segment Representing the The horizontal coordinate increment of the extreme center point within each periodic segment Representing the The vertical coordinate increment of the extreme center point within a periodic segment For the first The rate of change of local curvature of the extreme value trajectory within a periodic segment This represents the number of segments within a statistical period.

[0018] As a further aspect of the present invention, the step of obtaining the trajectory cooperative structure mapping map specifically includes:

[0019] S211: Call the amplitude path and phase shift path in the optical interference cooperative trajectory group, detect the time evolution trend of the amplitude path and phase shift path in the trajectory sequence between adjacent period groups, determine whether the change direction of the amplitude path and the change direction of the phase shift path are synchronized, and obtain the change direction synchronization index sequence.

[0020] S212: Based on the synchronization index sequence of the change direction, sort and calculate the synchronization index values ​​of multiple trajectory nodes in a continuous time segment, combine the difference magnitude between the synchronization index values ​​of adjacent trajectory nodes, filter the trajectory node clusters whose difference magnitude is less than the synchronization difference threshold, and mark the position of their respective period segment in the evolution sequence to obtain the distribution set of evolutionary homogeneous trajectory nodes.

[0021] S213: Based on the distribution set of the evolutionary homogeneous trajectory nodes, normalize the sequence position and synchronous change direction of the trajectory nodes in the amplitude path and phase offset path in the differentiated periodic segment, calculate the structural offset intensity value, and combine the periodic segment marking information to perform trajectory grouping and aggregation to generate a trajectory cooperative structure mapping diagram.

[0022] As a further aspect of the present invention, the step of obtaining the neural network recognition channel set specifically includes:

[0023] S311: Based on the characteristics of the marked trajectory clusters in the trajectory cooperative structure mapping diagram, the spatial coordinate sequence of the path trajectory in the trajectory set is divided into continuous segments, and a set of trajectory morphology factors including the path direction angle sequence and curvature change coefficient is extracted from the divided segments. By grouping and encoding the set of trajectory morphology factors according to the preset time window parameters, trajectory morphology encoding information is obtained.

[0024] S312: Call the trajectory morphology encoding information, input it into the multi-layer neural network recognition layer, and perform vector similarity retrieval on the trajectory label vectors of the path nodes in the neural network sample set. Quantify and score the cosine similarity between the retrieved label vectors and the input encoding matrix, and select the sample path nodes with the highest scores based on the similarity score ranking results to generate a matching path node index set.

[0025] S313: Based on the path tag status corresponding to the matching path node index set, perform clustering and aggregation processing on the path set with response status, and map the path set to the trajectory encoding and the mapping path of the neural network recognition layer node to generate the neural network recognition channel set.

[0026] As a further aspect of the present invention, the step of obtaining the structure diagram after path interference elimination specifically includes:

[0027] S411: Call the neural network to identify the interference pattern region corresponding to each path in the channel set, and perform coordinate positioning and brightness value extraction on the edge pixels of the light stripe in the interference pattern region in consecutive frames. Normalize the time series formed by the brightness values ​​of pixels at the same position in consecutive frames, and calculate the average disturbance amplitude based on the statistical value of brightness fluctuation amplitude between frames to obtain the edge pixel disturbance intensity sequence.

[0028] S412: Based on the edge pixel perturbation intensity sequence, the perturbation intensity value of the region corresponding to each path is compared with the path response time length item by item. By judging whether the perturbation intensity value exceeds the perturbation threshold corresponding to the response time length at each time point, it is determined whether there is a perturbation-dominated situation in the region where the path is located, and a path interference judgment label set is generated.

[0029] S413: Based on the path interference judgment label set, remove the paths marked as disturbance-dominant from the neural network recognition structure, reconstruct the connection sequence of the remaining paths in the original channel structure, and recombine the structure of the path nodes that were not removed to generate a structure diagram after path interference removal.

[0030] As a further aspect of the present invention, the method further includes step S5:

[0031] S5: Call the set of remaining identification paths in the structure diagram after path interference elimination, monitor the trend curve of the output probability changing with the input period in each path channel, match whether the fluctuation direction is consistent with the change of channel optical power, filter the identification path with the matching change direction, set the category node marked by the identification path as the identification output label of the real-time optical interference input, and generate the fiber-driven identification path output result.

[0032] The output results of the fiber-optic driven identification path include classification labels, matching path indexes, and periodic response trend parameters.

[0033] As a further aspect of the present invention, the step of obtaining the output result of the fiber optic drive identification path is specifically as follows:

[0034] S511: Call the set of remaining identification paths in the structure diagram after path interference elimination, extract the output probability of each identification path channel in a periodic segment within a continuous input period, construct a path output probability change curve based on an equally spaced time axis, and generate an output probability change trend set by combining the curve slope change value and the curve inflection point sequence.

[0035] S512: Based on the output probability change trend set, perform a direction consistency matching operation on the optical power change curve collected in the corresponding channel of each path, determine whether the rising or falling direction of each change segment in the trend direction sequence maintains the same label value as the corresponding optical power change segment, mark the path with consistent direction, extract the associated category node label, and obtain the direction consistency identification path label set.

[0036] S513: Based on the direction-consistent identification path label set, the category node label corresponding to the identification path is set as the identification output identifier under the real-time input cycle, and the identification path output identifiers are aggregated in order according to the identification path number to generate the fiber-driven identification path output result.

[0037] The fiber-driven neural network pattern recognition system is used to execute the above-described fiber-driven neural network pattern recognition method. The system includes:

[0038] The signal acquisition module acquires the optical interference signal data stream deployed at the optical fiber sensing node, detects the frequency stable section, optical power change trend and time sampling point density value in the optical wave interference signal, divides the data segment according to the periodic oscillation amplitude of the light wave and marks the group number according to each period, extracts the amplitude peak sequence, valley sequence and center trajectory point, records the extension path of the center trajectory in the period dimension, and establishes the optical interference cooperative trajectory group;

[0039] The trajectory construction module calls the amplitude path and phase offset path in the optical interference cooperative trajectory group, performs directional judgment on the path segment between each pair of adjacent period groups, extracts the displacement gradient coefficient and angle deviation ratio of adjacent paths in the time axis direction, marks the trajectory group number and clusters according to path density to obtain the trajectory cooperative structure mapping map.

[0040] The collaborative recognition module extracts the trajectory shape change slope sequence, peak spacing vector, and displacement fluctuation period value according to the number and trajectory shape of each group of collaborative trajectories in the trajectory collaborative structure mapping diagram. It forms a feature encoding matrix and inputs it into the neural network structure of the loaded sample path library. It then filters out path nodes with the same trend direction of phase offset peak and marks their corresponding response categories to generate a neural network recognition channel set.

[0041] The interference removal module calls the neural network to identify the channel set, collects the light stripe edge pixel sequence in each path region, detects the edge pixel gradient direction and pixel amplitude change rate in consecutive image frames, calculates the ratio of the pixel amplitude change rate in each path to its own path response time length, and if the ratio reflects that the disturbance trend is dominant, the path is marked as an unstable path and removed, generating a path interference removal structure diagram.

[0042] The path output module calls the structure diagram after path interference elimination, collects the output probability trend sequence and corresponding optical power value change rate of each path in a continuous period, calculates the direction consistency discrimination factor of the two, filters the paths with direction consistency, extracts the corresponding category nodes in the path as recognition output, and generates fiber-driven recognition path output results.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0044] In this invention, the periodic structural characteristics of optical interference signals are introduced to divide the input data into paths and trajectories. A co-evolutionary structure is established using the multidimensional variation trends of amplitude and phase, enhancing the ability to characterize the intrinsic connections between dynamic signal features. The recognition path is constructed through trajectory morphology encoding and combined with a neural network retrieval mechanism, effectively improving the recognition accuracy of evolution patterns. Combined with a dynamic detection mechanism of edge perturbation trends in optical interference patterns, unstable paths are eliminated in real time to suppress the interference of high-interference input on the recognition results, improving the stability and accuracy of recognition in high-noise environments. By analyzing the matching relationship between probability output and optical power fluctuations in the recognition path, the recognition output results are ensured to have stronger response consistency and classification effectiveness. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the workflow of the present invention;

[0046] Figure 2This is a flowchart illustrating the acquisition of the optical interference cooperative trajectory group in this invention.

[0047] Figure 3 This is a flowchart illustrating the process of obtaining the trajectory cooperative structure mapping map in this invention.

[0048] Figure 4 This is a flowchart illustrating the process of obtaining the neural network recognition channel set in this invention.

[0049] Figure 5 This is a flowchart illustrating the process of obtaining the structure diagram after path interference elimination in this invention.

[0050] Figure 6 This is a flowchart illustrating the process of obtaining the output results of the fiber optic drive identification path in this invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0053] Please see Figure 1 This invention provides a technical solution: a fiber-optic driven neural network pattern recognition method, comprising the following steps:

[0054] S1: Based on the optical interference signal data stream collected by the optical fiber sensing node, it is divided into continuous periodic groups according to the periodic oscillation characteristics of the light wave. The extreme center trajectory of the amplitude change in each group is extracted and the time path is recorded. The angle corresponding to the phase peak change point is extracted and a continuous offset trajectory is formed. A structural sequence group with amplitude path and phase path as elements is constructed to generate an optical interference cooperative trajectory group.

[0055] S2: Call the amplitude path and phase shift path in the optical interference cooperative trajectory group, detect whether the change direction of the two types of trajectories between adjacent period groups maintains a synchronous trend, mark the trajectory cluster with continuous unidirectional evolution behavior as cooperative evolution mode, classify the groups according to the continuity and synchronicity of the evolution paths between trajectories, and generate a trajectory cooperative structure mapping map.

[0056] S3: Based on the characteristics of the marked trajectory clusters in the trajectory cooperative structure mapping map, extract the morphological encoding of the path trajectory and input it into the neural network recognition layer. Retrieve path nodes with the same trajectory pattern label in the neural network sample set, mark the path set with response state as channels, and load the path set as the corresponding recognition path for real-time input to generate the neural network recognition channel set.

[0057] S4: Call the neural network to identify the interference pattern region corresponding to each path in the channel set, collect the state change information of the light stripe edge pixels in the interference pattern region in multiple frames of images, and detect the trend of disturbance feature change. If the interference is dominant, mark the path as an unstable channel, remove the path in the real-time recognition structure, and generate a structure diagram after path interference is eliminated.

[0058] S5: Call the set of remaining identification paths in the structure diagram after path interference elimination, monitor the trend curve of the output probability changing with the input period in each path channel, match whether the fluctuation direction is consistent with the change of channel optical power, filter the identification path with the matching change direction, set the category node marked by the identification path as the identification output label of the real-time optical interference input, and generate the fiber-driven identification path output result.

[0059] The optical interference cooperative trajectory group includes trajectory node association, trajectory periodic coherence information, and path synchronization offset features. The trajectory cooperative structure mapping diagram includes path clustering structure, cooperative label nodes, and mapping evolution trend features. The neural network recognition channel set includes encoded path set, neural response state, and target pattern channel index. The structure diagram after path interference elimination includes stable recognition path set, image perturbation shielding information, and channel label set. The fiber-driven recognition path output results include classification label, matching path index, and periodic response trend parameters.

[0060] Please see Figure 2 The specific steps for obtaining the optical interferometric cooperative trajectory set are as follows:

[0061] S111: Based on the optical interference signal data stream collected by the optical fiber sensing node, detect the boundary features of the oscillation period of the signal, extract the periodically adjacent extreme point pairs, calculate the difference between the mean and peak values ​​of the oscillation within the continuous period, classify them to form period intervals, record the period interval index and oscillation amplitude characteristic parameters of each group, and generate a periodic oscillation amplitude index group.

[0062] Within each sampling period, the boundary features of the oscillation period of the signal are extracted. The boundary of the oscillation period is defined by identifying the time nodes corresponding to two adjacent peaks in the optical interference intensity curve. For example, in a sampling sequence, if the optical interference signal has local maxima at 1.24s, 1.49s, and 1.73s, these three points are considered to constitute the boundary points of two oscillation periods. The extreme point pairs within each period are extracted one by one, and the difference between the maximum and minimum values ​​in each period segment is calculated as the oscillation amplitude. For example, if the maximum value is 3.76V and the minimum value is 2.05V in a certain period segment, the oscillation amplitude is 1.71V. The amplitudes of each period segment are arranged into an amplitude sequence in chronological order. The amplitude ranges of multiple period segments are aggregated and classified using a sliding window method to divide them into multiple period groups. Each group contains several continuous period segments. For example, if each group contains 10 period segments, and the total length of the amplitude sequence is 100 period segments, then 10 period groups are formed. The index number of each period and the corresponding oscillation amplitude are recorded in each group to generate a periodic oscillation amplitude index group.

[0063] S112: Call the periodic oscillation amplitude index group, filter the extreme value center trajectory within the periodic group, and extract continuous extreme points in the amplitude changes using the following formula:

[0064] ;

[0065] Calculate the amplitude offset trajectory transformation rate to obtain the trend of offset trajectory change;

[0066] in, Represents the amplitude offset trajectory change rate. Representing the Extreme amplitude values ​​within a periodic segment Representing the Extreme amplitude values ​​within a periodic segment Representing the The horizontal coordinate increment of the extreme center point within each periodic segment Representing the The vertical coordinate increment of the extreme center point within a periodic segment For the first The rate of change of local curvature of the extreme value trajectory within a periodic segment The number of segments in the statistical period;

[0067] Formula calculation logic: Based on the changes in the oscillation amplitude and the coordinate changes of the center position of the extreme points within multiple period segments, for each period segment... Calculate the extreme value amplitude Compared with the extreme value amplitude of the previous period The difference between Then, the difference is squared to reflect the intensity of the amplitude change in the continuous period segment, and the increment of the center position of the extreme point of the period segment in the horizontal and vertical directions is calculated respectively. Then, the squares of the two increments are added together to obtain the squared distance of the spatial position trajectory change. The product of the above two items is then divided by... ,in This represents the rate of change of local curvature within the current period segment, reflecting the complexity of trajectory changes. The entire expression... Calculate the average over each period to obtain... It represents an average indicator that shows the coordinated changes in oscillation amplitude and spatial location across multiple time periods.

[0068] The amplitude offset trajectory transformation rate is used to measure the comprehensive degree of the spatial displacement of the extreme point and the change of oscillation intensity during the change of oscillation amplitude with the period. It also considers the square of the change of oscillation amplitude in adjacent periods and the two-dimensional coordinate movement of the trajectory center of the extreme point. The indicator reflects the superposition intensity of trajectory fluctuation and amplitude difference in the entire period sequence by dividing by the local curvature change adjustment factor and taking the average.

[0069] Parameter meaning and calculation process:

[0070] This represents the rate of change of local curvature within the current period segment. Its value is obtained by fitting the second derivative of the trajectory curve using a three-point difference method and dividing by the square of the first derivative to the power of 1.5. If we define the three consecutive center points of a certain period segment as... , , ;

[0071] Then the calculation yields , ;

[0072] The amplitude values ​​are respectively , The rate of change of curvature was calculated to be Then, substituting into the formula and calculating item by item, the results are as follows:

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] Suppose the entire periodic group contains Given a period of 9 terms, the average of the first 9 terms is 0.00875. Therefore:

[0078] ;

[0079] The results indicate that there are significant changes in the oscillation amplitude and extreme value trajectory of the current period group, suggesting that the local deformation of the current period group is quite obvious, which is suitable for detecting the rotation direction transition point;

[0080] Table 1: Parameter Table of Trajectory Transformation Rate in Periodic Segments

[0081]

[0082] As shown in Table 1, all parameters are selected within the physically measurable range. The trajectory transformation rate is ultimately obtained from the average value of each period segment, which is convenient for subsequent use in position recognition.

[0083] S113: Based on the trend of the offset trajectory change, identify the phase angle point corresponding to the peak transfer point, record the position sequence and angle change sequence in the continuous period group, construct the corresponding trajectory segment direction vector group, and combine it with the original amplitude center trajectory sequence to generate the optical interference cooperative trajectory group.

[0084] Within each periodic group, curvature abrupt change points are extracted from the continuous trajectory change rate sequence and used as candidate direction transition points. Nodes with a change rate exceeding a set threshold are identified; for example, a direction transition threshold of 0.012. When the change rate of a trajectory in the periodic group exceeds the threshold (e.g., the change rate of period segment 8 is 0.0142), the curvature abrupt change point is determined as a candidate direction transition point. The position index of each node in the global periodic group is recorded sequentially, and the angle between this index and the displacement direction of adjacent points is calculated. Let the current position be... The previous node is The next node is Then the forward vector is The backward vector is According to the formula for the angle between two vectors:

[0085] ;

[0086] We can obtain:

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] Based on the judgment criteria, when the included angle is greater than 8° and the transformation rate is greater than the threshold, it is identified as a valid direction change point. The valid direction change point is inserted into the trajectory direction vector sequence and paired with the original center trajectory to construct an optical interference cooperative trajectory group for offset identification and analysis.

[0092] Please see Figure 3 The specific steps for obtaining the trajectory cooperative structure mapping graph are as follows:

[0093] S211: Call the amplitude path and phase shift path in the optical interference cooperative trajectory group, detect the time evolution trend of the amplitude path and phase shift path in the trajectory sequence between adjacent period groups, determine whether the change direction of the amplitude path and the change direction of the phase shift path are synchronized, and obtain the sequence of change direction synchronization index.

[0094] Extract the timestamp sequence corresponding to the trajectory nodes between adjacent period groups, and then perform node... Extract the time position value in the amplitude path from the trajectory dataset and set it as... and the corresponding time position in the phase offset path. Establish two time-series vectors of equal length, and sequentially traverse the amplitude path change direction of adjacent trajectory nodes in each period group. Direction of phase offset path change The system checks if the symbols are consistent. If they are consistent, it considers the system synchronized and records it as 1; otherwise, it records it as 0. The synchronization value is then marked as the value of the current position in the synchronization index sequence. For example, when... , When both change in the same direction, the index value is recorded as 1. , When the sign is different, the index value is recorded as 0. The synchronicity index sequence within the entire period is stored and a synchronicity vector of length T is constructed. The sequence of indicators for the synchronicity of the direction of change is obtained, and the execution process is shown in Table 2.

[0095] Table 2: Trajectory Node Change Direction and Synchronization Indicators

[0096]

[0097] As shown in Table 2, by comparing the changing trends of each node under different paths, the synchronization state is determined and a complete synchronization sequence is formed for use.

[0098] S212: Based on the synchronicity index sequence of the changing direction, the synchronicity index values ​​of multiple trajectory nodes in a continuous time segment are sorted and calculated. Combined with the difference magnitude between the synchronicity index values ​​of adjacent trajectory nodes, the trajectory node clusters with difference magnitude less than the synchronicity difference threshold are screened and marked with their respective period segment positions in the evolution sequence to obtain the distribution set of evolutionary homogeneous trajectory nodes.

[0099] Extract the synchronization index values ​​of multiple trajectory nodes within consecutive time segments in adjacent period groups. For example, in a period segment with T=5, let the node synchronization sequence be {1, 1, 0, 1, 0}. Calculate the index value difference under a fixed window sliding condition. For example, with a window length of 3, compare the differences between [1, 1, 0] and [1, 0, 1], and between [1, 0, 1] and [0, 1, 0]. Define the difference as the XOR sum of the synchronization indices at corresponding positions within the window. That is, after XORing the first window with the second window, we get {0, 1, 1}, and summing them gives a difference of 2. Repeat this comparison to obtain the difference sequence D for the entire segment, and then filter the results. Starting positions with a difference value not exceeding 1 form a candidate trajectory node cluster set. For example, in the above sequence, [1, 1, 0] and [1, 0, 1] have a difference value of 2 and are eliminated, and [1, 0, 1] and [0, 1, 0] have a difference value of 3 and are also eliminated. Only continuous segments that meet the conditions are retained. During the execution process, the difference threshold is set to 1, and the reasonable range is between 0 and 3. Based on the physical stability requirements in trajectory changes, the average difference value is 1.4 obtained by statistically analyzing 10 sets of sample sequences. Setting the threshold to 1 is operational, and segments falling outside the threshold are automatically eliminated to obtain the distribution set of evolutionary unidirectional trajectory nodes.

[0100] S213: Based on the distribution set of evolutionary homogeneous trajectory nodes, the sequence positions and synchronous change directions of trajectory nodes in the differential periodic segments within the amplitude path and phase shift path are normalized using the following formula:

[0101] ;

[0102] Calculate the structural offset strength value, and combine it with the period segment marking information to perform trajectory grouping and aggregation, and generate a trajectory cooperative structure mapping map;

[0103] in, Representing the trajectory With trajectory Structural offset strength values ​​between Representing the trajectory At any moment The path-normalized position value, Representing the trajectory At any moment The path-normalized position value, Representing the trajectory At any moment Direction normalized offset value, Representing the trajectory At any moment Direction normalized offset value, This represents the total number of time steps within a period.

[0104] Formula calculation logic: Based on the time-normalized difference between the amplitude path and the phase path of each trajectory node within the period. The difference represents the degree of deviation in the trajectory positions of the two paths at the same point in time. The difference value is then multiplied by a directional offset amplification factor. , and These are the normalized offsets of the amplitude path and phase path in terms of direction, respectively. The square root operation is used to enhance the sensitivity of the directional offset in the overall difference. The summation and averaging of the products at each time point are normalized to the overall structural offset intensity within the period segment. The larger the value, the more significant the trajectory difference between the amplitude path and the phase path in space and direction.

[0105] The structural offset intensity value reflects the overall degree of deviation between the amplitude path and the phase path in terms of trajectory space and directional changes. It is a quantitative indicator for measuring the coordination of the two paths. The value comprehensively considers the time-normalized position difference and directional offset amplitude, which can effectively reveal the inconsistency in the trajectory structure. The closer the value is to 0, the more consistent the two paths are in the same period segment. The larger the value, the more the structural difference intensifies.

[0106] The formula parameters are explained as follows:

[0107] This represents the number of trajectory nodes within the period, which is 10 in the actual sample.

[0108] , To represent the normalized trajectory positions of the amplitude path and the phase path at time t, respectively, in seconds, normalized to [0, 1];

[0109] , This represents the direction normalization offset value, in radians, and its range is [0, 1].

[0110] Absolute value part The square root represents the time difference between the two paths. Indicates the directional offset amplification term;

[0111] Now, let T=5, and set the following parameters in sequence:

[0112] Table 3: Parameter setting values ​​at different times

[0113]

[0114] The formula then expands as follows:

[0115] ;

[0116] The results show that the structural offset intensity value is 0.0279, which is in the low range of the normalized intensity range [0, 1]. Based on the structural offset stability, the benchmark value is set to 0.05. The value is lower than the benchmark value, which indicates that the structural differences between trajectories are small and will be used for subsequent tagging information aggregation operations.

[0117] Please see Figure 4 The specific steps for obtaining the neural network recognition channel set are as follows:

[0118] S311: Based on the characteristics of the marked trajectory clusters in the trajectory cooperative structure mapping map, the spatial coordinate sequence of the path trajectory in the trajectory set is divided into continuous segments, and a set of trajectory morphology factors including the path direction angle sequence and curvature change coefficient is extracted from the divided segments. By grouping and encoding the set of trajectory morphology factors according to the preset time window parameters, trajectory morphology encoding information is obtained.

[0119] It is necessary to extract the coordinate sequence of the path trajectory from the trajectory data. The trajectory is a set of geographical points arranged in chronological order. In practice, the entire path is divided into segments based on the changes in spatial distance between trajectory points. Suppose a vehicle travels from the city center to the suburbs. The trajectory shows that it will experience obvious path turning points during its journey, such as turning from a straight road into a winding mountain road. The distance or angle between the coordinate points will change significantly. Based on the changes, the trajectory is divided into several segments. The trajectory point sequence in each continuous segment is analyzed to extract the path orientation change information and the curvature of the trajectory trend in each segment. The morphological factor is considered an important feature of the segment. For example, a route with frequent turns in urban roads will show multiple changes in direction and high-frequency curvature in the morphological factor. On the other hand, a long straight path on a highway has small changes in direction and stable curvature. The features are uniformly encoded and organized into fixed time windows. For example, the trajectory morphological features within every 5 seconds are combined into a coding group to obtain the trajectory morphological coding information.

[0120] S312: Call the trajectory morphology encoding information, input it into the multi-layer neural network recognition layer, and perform vector similarity retrieval on the trajectory label vectors of the path nodes in the neural network sample set. Quantify and score the cosine similarity between the retrieved label vectors and the input encoding matrix, and select the top-ranked sample path nodes based on the similarity score ranking results to generate a matching path node index set.

[0121] The encoded vector is input into the constructed neural network recognition layer for analysis. During this process, the morphological information of each trajectory is converted into input data in a unified format. A sequence of multiple values ​​represents the morphological structure of a path. After entering the recognition layer, the path node labels already labeled in the training sample set are called, and the label vectors are compared with the current input data one by one. In order to achieve accurate matching, a similarity calculation method is used to measure the degree of consistency between the input trajectory and the sample path in terms of overall shape. For example, a path that makes multiple turns near a city intersection will be matched with sample paths with similar trends. The results are compared and ranked, and several sample path nodes with the highest similarity ranking are selected as candidates. The candidate results are the set of paths considered to be closest to the current input trajectory, which is used for path clustering analysis and neural network recognition channel construction. It can be executed quickly in large-scale trajectory databases and is suitable for high-frequency trajectory recognition application scenarios, such as real-time path recognition and path behavior prediction in intelligent transportation, generating a matching path node index set.

[0122] S313: Based on the path tag status corresponding to the matching path node index set, perform clustering and aggregation processing on the path set with response status, and map the path set to the trajectory encoding and the mapping path of the neural network recognition layer node to generate the neural network recognition channel set;

[0123] The system queries the path status information corresponding to each index node and uses it as the basis for clustering. Taking autonomous driving as an example, assuming that the trajectory status of multiple nodes is marked as "left turn behavior", the paths with the same behavior label are aggregated into a group. During the aggregation process, not only is the consistency of behavior labels considered, but also the similarity of morphological features between the paths is taken into account. The paths are then filtered and merged again. When the clustering results are stable, each cluster group will form a set of paths with typical trajectory morphology. According to the morphological code of the path in the cluster group, it is mapped to a specific node in the neural network structure. For example, if a cluster group represents the path of an urban elevated bridge section, the corresponding morphological code is bound to a specific recognition channel in the neural network, enabling the rapid identification of similar paths. The formation of the recognition channel is based on the correspondence between trajectory morphology, path status, and neural network nodes. It is a key component in the intelligent path recognition system and is suitable for large-scale traffic path classification and traffic flow prediction scenarios, generating a set of neural network recognition channels.

[0124] Please see Figure 5 The specific steps for obtaining the structure diagram after path interference elimination are as follows:

[0125] S411: Call the neural network to identify the interference pattern region corresponding to each path in the channel set, and perform coordinate positioning and brightness value extraction on the edge pixels of the light stripe in the interference pattern region in multiple consecutive frames. Normalize the time series formed by the brightness values ​​of pixels at the same position in consecutive frames, and calculate the average disturbance amplitude based on the statistical value of brightness fluctuation amplitude between frames to obtain the edge pixel disturbance intensity sequence.

[0126] Interference analysis is required for the image region corresponding to each path. Specifically, this involves processing multiple frames of images of the interference pattern region. Based on the mapping relationship between each path and the image frame, the image region covered by the path is locked in the video image. Then, the edges formed by the interference fringes, especially the boundary pixels of the light fringes, are identified in the region. The edges are represented by the locations of rapid brightness changes. The pixels are located using an image grayscale gradient scanning algorithm. Taking the interference pattern generated by a car passing over a bridge as an example, the edge light fringes are located in the area where the wheel and the bridge meet. The position and brightness value of the pixels are recorded from the first frame. When the image enters the next frame, the same pixel position is located again and the new brightness value is recorded. The entire image sequence is processed in sequence. After the image extraction is completed, the brightness values ​​of each edge pixel in multiple frames are uniformly arranged into a time series and normalized to eliminate the influence of illumination changes. The inter-frame variation amplitude between normalized brightness values ​​is statistically analyzed to obtain the perturbation amplitude of a certain pixel over time and generate a path interference judgment label set.

[0127] S412: Based on the edge pixel perturbation intensity sequence, the perturbation intensity value of the region corresponding to each path is compared with the path response time length item by item. By judging whether the perturbation intensity value exceeds the perturbation threshold corresponding to the response time length at each time point, it is determined whether there is a perturbation-dominated situation in the region where the path is located, and a path interference judgment label set is generated.

[0128] Each path's image region is analyzed item by item. In practice, for each disturbance intensity value, the response time parameter corresponding to the path is retrieved. This parameter is set during the path behavior modeling stage using time or event reaction time. For example, in automatically detecting railway curves, a path's response time is 4 seconds, indicating that stable data is needed within the time range for judgment. The disturbance intensity of each frame is compared to see if it remains above the set threshold. If the continuous brightness disturbance values ​​are all higher than the disturbance threshold for the corresponding time period, it is considered that the path region has continuous disturbances caused by external intervention. Taking the appearance of high brightness fluctuations in 5 consecutive frames of a path as an example, it is judged that the disturbance trend is obvious and continuous, exceeding the standard corresponding to the response time, which is considered as interference dominated by external factors. In this process, the setting of the disturbance threshold needs to be determined in combination with the noise baseline value in the image sample, the common disturbance amplitude distribution, and the stable range of edge vibration. For example, if the overall interference value of images taken at night is high, the threshold setting range needs to be increased to adapt to the lighting conditions. The same judgment is performed on the path region in turn, and the judgment result of whether each path is dominated by disturbance is output, generating a path interference judgment label set.

[0129] S413: Based on the path interference judgment label set, remove the paths marked as perturbation-dominant from the neural network recognition structure, reconstruct the connection sequence of the remaining paths in the original channel structure, and recombine the structure of the path nodes that were not removed to generate a structure diagram after path interference removal.

[0130] The removal operation on paths marked as having significant disturbances is to eliminate the interference caused by path nodes severely affected by external disturbances to the overall recognition structure. The removal operation traverses the original channel connection structure of the neural network, identifies the connection nodes and sub-channels associated with the marked path nodes. If a main path node is connected to multiple lower-level branch nodes in the recognition structure, the structure will be located and the connection path including the main node will be removed. The connection sequence of the remaining unaffected path nodes will be reconstructed. Specifically, the order of connection relationships between nodes will be re-evaluated, and the nodes will be reordered and combined according to the connection logic and dependency rules in the recognition task to ensure the continuity and effectiveness of the path recognition process. In the automatic trajectory recognition task, if a trajectory is removed due to excessive edge interference, an attempt will be made to find a replacement node with a similar trajectory shape, and the network structure will be adjusted according to the connection information to ensure that the recognition channel still has integrity. The reconnected path structure information is uniformly organized to generate a structure diagram after path interference removal.

[0131] Please see Figure 6 The specific steps for obtaining the output results of the fiber optic drive identification path are as follows:

[0132] S511: Call the set of remaining recognition paths in the structure diagram after path interference elimination, extract the output probability of each recognition path channel in a periodic segment within a continuous input period, and construct a path output probability change curve based on an equally spaced time axis. Combine the curve slope change value and the curve inflection point sequence to generate a set of output probability change trends.

[0133] When analyzing the set of identified paths retained in the structure diagram after eliminating path interference, it is necessary to extract the output probability data of each path channel within a continuous input cycle. The output probability results of each path in the current cycle are retrieved from the identification model and organized into a list in chronological order. The entire time axis is evenly divided according to the set time sampling interval. The data in each segment is regarded as a data block of an independent cycle for easy analysis. Taking a real-time monitoring cycle of 10 seconds as an example, it is divided into 5 segments, each segment with a length of 2 seconds. The output probability of each path is extracted and organized into a time series. With time as the horizontal axis and output probability as the vertical axis, the periodic output probability points are connected to form a continuous curve to describe the dynamic change of the path output probability over time. In order to reveal the trend of change, the slope of the curve in each time period is analyzed. The numerical changes between adjacent points are used to determine whether the curve is rising, falling, or remaining stable. At the same time, the positions in the curve where there is a sudden change in direction or a reversal of trend are identified as inflection points. The slope change of the curve, the inflection point sequence, and the corresponding time points are combined to generate a set of output probability change trends.

[0134] S512: Based on the output probability change trend set, perform a direction consistency matching operation on the optical power change curve collected in the corresponding channel of each path, determine whether the rising or falling direction of each change segment in the trend direction sequence maintains the same label value as the corresponding optical power change segment, mark the path with consistent direction, extract the associated category node label, and obtain the direction consistency identification path label set.

[0135] The optical power change curves of the corresponding paths in the trend set are compared with those in the optical acquisition. The optical power change curves are obtained from the real-time sampling results of the fiber optic sensor and reflect the actual optical energy transmission of the monitored physical path within a certain period. The acquired optical power curves are segmented along the same time axis so that each segment corresponds one-to-one with the aforementioned trend segment of output probability. To achieve consistency matching, the direction of the output probability change trend of each segment is extracted, which can be represented by two directional indicators: "upward" or "downward". Then, the change direction within the same time period in the optical power curve is extracted accordingly. If the two directional indicators are the same, the trend is considered to have directional consistency. In specific application scenarios, for example, if a path shows a continuous upward trend in the optical power curve and the output probability also gradually increases within the same time period, the path is considered to meet the directional consistency condition. Each path that meets the condition is labeled, and the category node labels associated with the path in the neural network structure are extracted to obtain the directional consistency identification path label set.

[0136] S513: Based on the direction-consistent identification path label set, the category node label corresponding to the identification path is set as the identification output identifier under the real-time input cycle, and the identification path is ordered according to the identification path number. By aggregating the identification path output identifiers within the real-time cycle, the fiber-driven identification path output result is generated.

[0137] The system processes the recognition output within the current real-time input cycle. Each path has been mapped to a category node label during recognition. The labels in the set of paths with consistent direction can be directly assigned as the recognition results for this cycle. During operation, the path labels are sorted and registered one by one according to the numbering order of the recognition paths to ensure a stable output structure. Taking an industrial inspection scenario as an example, when multiple pipeline paths are identified with path nodes that have consistent signal responses, the labels are set as valid recognition paths in the current cycle. The remaining paths that do not meet the consistency condition are not included in the current results. The labels of the valid recognition paths in this cycle are aggregated to construct a complete cycle output data unit. Modules such as path on / off control and signal feedback execute control strategies to ensure the real-time performance and continuity of the recognition link and generate fiber-optic driven recognition path output results.

[0138] The fiber-optic driven neural network pattern recognition system is used to execute the above-described fiber-optic driven neural network pattern recognition method. The system includes:

[0139] The signal acquisition module acquires the optical interference signal data stream deployed at the optical fiber sensing node, detects the frequency stable section, optical power change trend and time sampling point density value in the optical wave interference signal, divides the data segment according to the periodic oscillation amplitude of the light wave and marks the group number according to each period, extracts the amplitude peak sequence, valley sequence and center trajectory point, records the extension path of the center trajectory in the period dimension, and establishes the optical interference cooperative trajectory group;

[0140] The trajectory construction module calls the amplitude path and phase offset path in the optical interference cooperative trajectory group, performs directional judgment on the path segment between each pair of adjacent period groups, extracts the displacement gradient coefficient and angle deviation ratio of adjacent paths in the time axis direction, marks the trajectory group number and clusters according to path density to obtain the trajectory cooperative structure mapping map.

[0141] The collaborative recognition module extracts the trajectory shape change slope sequence, peak spacing vector, and displacement fluctuation period value based on the number and trajectory shape of each group of collaborative trajectories in the trajectory collaborative structure mapping diagram. It forms a feature encoding matrix and inputs it into the neural network structure of the loaded sample path library. It then filters out path nodes with the same trend direction of phase offset peak and marks their corresponding response categories to generate a neural network recognition channel set.

[0142] The interference removal module calls the neural network to identify the channel set, collects the edge pixel sequence of light stripes in each path region, detects the gradient direction of edge pixels and the rate of change of pixel amplitude in consecutive image frames, calculates the ratio of the rate of change of pixel amplitude in each path to the length of its own path response time. If the ratio reflects that the disturbance trend is dominant, the path is marked as an unstable path and removed, generating a structure diagram after path interference removal.

[0143] The path output module calls the structure diagram after path interference elimination, collects the output probability trend sequence and corresponding optical power value change rate of each path in a continuous period, calculates the directional consistency discrimination factor of the two, filters the paths with directional consistency, extracts the corresponding category nodes in the path as recognition output, and generates fiber-driven recognition path output results.

[0144] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A fiber-optic driven neural network pattern recognition method, characterized in that, Includes the following steps: S1: Based on the optical interference signal data stream collected by the optical fiber sensing node, it is divided into continuous periodic groups according to the periodic oscillation characteristics of the light wave. The extreme value center trajectory in the amplitude change within each group is extracted and the time sequence path is recorded to generate an optical interference cooperative trajectory group. S2: Call the amplitude path and phase offset path in the optical interference cooperative trajectory group, detect whether the change direction of the two types of trajectories between adjacent period groups maintains a synchronous trend, mark the trajectory cluster with continuous unidirectional evolution behavior as cooperative evolution mode, and generate a trajectory cooperative structure mapping map. The optical interference cooperative trajectory group includes trajectory node association, trajectory periodic continuity information, and path synchronization offset features. The trajectory cooperative structure mapping diagram includes path clustering structure, cooperative label nodes, and mapping evolution trend features. The neural network recognition channel set includes encoded path set, neural response state, and target pattern channel index. The structure diagram after path interference elimination includes stable recognition path set, image perturbation shielding information, and channel label set. S3: Based on the trajectory cluster features marked in the trajectory cooperative structure mapping map, extract the morphological code of the path trajectory and input it into the neural network recognition layer, retrieve path nodes with the same trajectory pattern label in the neural network sample set, and generate a neural network recognition channel set. S4: Call the neural network to identify the channel set, collect the state change information of the edge pixels of the light stripes in the interference pattern area in multiple frames of images, detect the trend of disturbance feature changes, and generate a structural map after path interference is eliminated; It also includes step S5: S5: Call the set of remaining identification paths in the structure diagram after path interference elimination, monitor the trend curve of the output probability changing with the input period in each path channel, match whether the fluctuation direction is consistent with the change of channel optical power, filter the identification path with the matching change direction, set the category node marked by the identification path as the identification output label of the real-time optical interference input, and generate the fiber-driven identification path output result. The output results of the fiber-optic driven identification path include classification labels, matching path indexes, and periodic response trend parameters.

2. The fiber-driven neural network pattern recognition method according to claim 1, characterized in that, The specific steps for obtaining the optical interferometric cooperative trajectory set are as follows: S111: Based on the optical interference signal data stream collected by the optical fiber sensing node, detect the boundary features of the oscillation period of the signal, extract the periodically adjacent extreme point pairs, calculate the difference between the mean and peak values ​​of the oscillation within the continuous period, classify them to form period intervals, record the period interval index and oscillation amplitude characteristic parameters of each group, and generate a periodic oscillation amplitude index group. S112: Call the periodic oscillation amplitude index group, filter the extreme center trajectory within the periodic group, extract continuous extreme points in amplitude changes, calculate the amplitude offset trajectory transformation rate, and obtain the offset trajectory change trend. S113: Based on the trend of the offset trajectory change, identify the phase angle point corresponding to the peak transfer point, record the position sequence and angle change sequence in the continuous period group, construct the corresponding trajectory segment direction vector group, and combine it with the original amplitude center trajectory sequence to generate the optical interference cooperative trajectory group.

3. The fiber-driven neural network pattern recognition method according to claim 2, characterized in that, The amplitude offset trajectory transformation rate is calculated using the following formula: ； in, Represents the amplitude offset trajectory change rate. Representing the Extreme amplitude values ​​within a periodic segment Representing the Extreme amplitude values ​​within a periodic segment Representing the The horizontal coordinate increment of the extreme center point within each periodic segment Representing the The vertical coordinate increment of the extreme center point within a periodic segment For the first The rate of change of local curvature of the extreme value trajectory within a periodic segment This represents the number of segments within a statistical period.

4. The fiber-driven neural network pattern recognition method according to claim 2, characterized in that, The specific steps for obtaining the trajectory cooperative structure mapping map are as follows: S211: Call the amplitude path and phase shift path in the optical interference cooperative trajectory group, detect the time evolution trend of the amplitude path and phase shift path in the trajectory sequence between adjacent period groups, determine whether the change direction of the amplitude path and the change direction of the phase shift path are synchronized, and obtain the change direction synchronization index sequence. S212: Based on the synchronization index sequence of the change direction, sort and calculate the synchronization index values ​​of multiple trajectory nodes in a continuous time segment, combine the difference magnitude between the synchronization index values ​​of adjacent trajectory nodes, filter the trajectory node clusters whose difference magnitude is less than the synchronization difference threshold, and mark the position of their respective period segment in the evolution sequence to obtain the distribution set of evolutionary homogeneous trajectory nodes. S213: Based on the distribution set of the evolutionary homogeneous trajectory nodes, normalize the sequence position and synchronous change direction of the trajectory nodes in the amplitude path and phase offset path in the differentiated periodic segment, calculate the structural offset intensity value, and combine the periodic segment marking information to perform trajectory grouping and aggregation to generate a trajectory cooperative structure mapping diagram.

5. The fiber-driven neural network pattern recognition method according to claim 4, characterized in that, The specific steps for obtaining the neural network identification channel set are as follows: S311: Based on the characteristics of the marked trajectory clusters in the trajectory cooperative structure mapping diagram, the spatial coordinate sequence of the path trajectory in the trajectory set is divided into continuous segments, and a set of trajectory morphology factors including the path direction angle sequence and curvature change coefficient is extracted from the divided segments. By grouping and encoding the set of trajectory morphology factors according to the preset time window parameters, trajectory morphology encoding information is obtained. S312: Call the trajectory morphology encoding information, input it into the multi-layer neural network recognition layer, and perform vector similarity retrieval on the trajectory label vectors of the path nodes in the neural network sample set. Quantify and score the cosine similarity between the retrieved label vectors and the input encoding matrix, and select the sample path nodes with the highest scores based on the similarity score ranking results to generate a matching path node index set. S313: Based on the path tag status corresponding to the matching path node index set, perform clustering and aggregation processing on the path set with response status, and map the path set to the trajectory encoding and the mapping path of the neural network recognition layer node to generate the neural network recognition channel set.

6. The fiber-driven neural network pattern recognition method according to claim 5, characterized in that, The specific steps for obtaining the structure diagram after path interference elimination are as follows: S411: Call the neural network to identify the interference pattern region corresponding to each path in the channel set, and perform coordinate positioning and brightness value extraction on the edge pixels of the light stripe in the interference pattern region in consecutive frames. Normalize the time series formed by the brightness values ​​of pixels at the same position in consecutive frames, and calculate the average disturbance amplitude based on the statistical value of brightness fluctuation amplitude between frames to obtain the edge pixel disturbance intensity sequence. S412: Based on the edge pixel perturbation intensity sequence, the perturbation intensity value of the region corresponding to each path is compared with the path response time length item by item. By judging whether the perturbation intensity value exceeds the perturbation threshold corresponding to the response time length at each time point, it is determined whether there is a perturbation-dominated situation in the region where the path is located, and a path interference judgment label set is generated. S413: Based on the path interference judgment label set, remove the paths marked as disturbance-dominant from the neural network recognition structure, reconstruct the connection sequence of the remaining paths in the original channel structure, and recombine the structure of the path nodes that were not removed to generate a structure diagram after path interference removal.

7. The fiber-driven neural network pattern recognition method according to claim 1, characterized in that, The specific steps for obtaining the output result of the fiber optic drive identification path are as follows: S511: Call the set of remaining identification paths in the structure diagram after path interference elimination, extract the output probability of each identification path channel in a periodic segment within a continuous input period, construct a path output probability change curve based on an equally spaced time axis, and generate an output probability change trend set by combining the curve slope change value and the curve inflection point sequence. S512: Based on the output probability change trend set, perform a direction consistency matching operation on the optical power change curve collected in the corresponding channel of each path, determine whether the rising or falling direction of each change segment in the trend direction sequence maintains the same label value as the corresponding optical power change segment, mark the path with consistent direction, extract the associated category node label, and obtain the direction consistency identification path label set. S513: Based on the direction-consistent identification path label set, the category node label corresponding to the identification path is set as the identification output identifier under the real-time input cycle, and the identification path output identifiers are aggregated in order according to the identification path number to generate the fiber-driven identification path output result.

8. A fiber-optic driven neural network pattern recognition system, characterized in that, The system is used to implement the fiber-driven neural network pattern recognition method according to any one of claims 1-7, the system comprising: The signal acquisition module acquires the optical interference signal data stream deployed at the optical fiber sensing node, detects the frequency stable section, optical power change trend and time sampling point density value in the optical wave interference signal, divides the data segment according to the periodic oscillation amplitude of the light wave and marks the group number according to each period, extracts the amplitude peak sequence, valley sequence and center trajectory point, records the extension path of the center trajectory in the period dimension, and establishes the optical interference cooperative trajectory group; The trajectory construction module calls the amplitude path and phase offset path in the optical interference cooperative trajectory group, performs directional judgment on the path segment between each pair of adjacent period groups, extracts the displacement gradient coefficient and angle deviation ratio of adjacent paths in the time axis direction, marks the trajectory group number and clusters according to path density to obtain the trajectory cooperative structure mapping map. The collaborative recognition module extracts the trajectory shape change slope sequence, peak spacing vector, and displacement fluctuation period value according to the number and trajectory shape of each group of collaborative trajectories in the trajectory collaborative structure mapping diagram. It forms a feature encoding matrix and inputs it into the neural network structure of the loaded sample path library. It then filters out path nodes with the same trend direction of phase offset peak and marks their corresponding response categories to generate a neural network recognition channel set. The interference removal module calls the neural network to identify the channel set, collects the light stripe edge pixel sequence in each path region, detects the edge pixel gradient direction and pixel amplitude change rate in consecutive image frames, calculates the ratio of the pixel amplitude change rate in each path to its own path response time length, and if the ratio reflects that the disturbance trend is dominant, the path is marked as an unstable path and removed, generating a path interference removal structure diagram. The path output module calls the structure diagram after path interference elimination, collects the output probability trend sequence and corresponding optical power value change rate of each path in a continuous period, calculates the direction consistency discrimination factor of the two, filters the paths with direction consistency, extracts the corresponding category nodes in the path as recognition output, and generates fiber-driven recognition path output results.

Images