Neuromorphic visual target classification method based on spiking neural network

Abstract

The invention relates to the technical field of visual target classification, in particular to a neuromorphic visual target classification method based on a spiking neural network, which comprises the following steps: collecting spiking neuron excitation data in a time window, detecting the pulse trigger frequency of target features, analyzing the feature correlation fluctuation condition of visual target classification, and classifying the visual targets. And identifying a feature change trend, screening continuously changing features as stable features, and generating a feature change classification result. According to the method, by adjusting the nerve cell cross-layer connection path and the synaptic transmission capability and optimizing stable and unstable features, the visual target classification efficiency and accuracy are further improved, especially the performance in a dynamic environment, and the visual target classification efficiency and accuracy are improved based on statistical analysis and adjustment of pulse time. And the recognition capability of the pulse neural network on the time sensitive target is enhanced, so that the time alignment pulse data can reflect the real-time change of visual input more accurately, and the target recognition and classification effect in a dynamic scene is remarkably optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of visual target classification, and particularly to a neuromorphic visual target classification method based on a spiking neural network. Background Art

[0002] The technical field of visual target classification is an important part of computer vision, mainly focusing on how to enable a computer to understand and classify different objects in an image through image recognition technology. This field involves a variety of technologies such as image processing, feature extraction, and machine learning algorithms to achieve automatic recognition and classification of objects, scenes, or activities. Specific technologies include image preprocessing, feature detection, pattern recognition, and the design and optimization of classifiers, aiming to improve the accuracy and efficiency of classification through technical means.

[0003] Among them, the neuromorphic visual target classification method based on a spiking neural network refers to using a spiking neural network to classify visual targets. This method mimics the processing mode of the biological nervous system and processes and transmits information through temporal spike activities. Technical matters cover the use of specific neuromorphic hardware or simulation software for the generation and transmission of spike signals and the target recognition process based on spike signals. This method is particularly suitable for processing visual data that is difficult to classify by traditional methods due to its temporal dynamic characteristics, and completes target classification by designing specific network structures and learning algorithms.

[0004] Existing visual target classification technologies mainly rely on traditional image processing and machine learning algorithms, but they are inadequate when dealing with dynamic or time-sensitive visual data, with insufficient sensitivity to time changes and difficulty in effectively capturing and utilizing the temporal dynamic information in visual data, especially in rapidly changing scenarios. In the feature extraction and classification processes, traditional methods ignore the analysis of feature stability and instability, resulting in poor recognition accuracy and processing efficiency in complex environments. The deficiencies limit their widespread deployment in practical applications, especially in application scenarios that require a highly dynamic response, such as autonomous driving and real-time monitoring. Summary of the Invention

[0005] To solve the technical problems existing in the prior art, such as being inadequate when dealing with dynamic or time-sensitive visual data, having insufficient sensitivity to time changes, and being difficult to effectively capture and utilize the temporal dynamic information in visual data, especially in rapidly changing scenarios, and traditional methods ignoring the analysis of feature stability and instability in the feature extraction and classification processes, resulting in poor recognition accuracy and processing efficiency in complex environments, the embodiments of the present invention provide a neuromorphic visual target classification method based on a spiking neural network. The technical solution is as follows: On the one hand, a neuromorphic visual target classification method based on a spiking neural network is provided, and the method includes: S1: Collect the spike neuron firing data within the acquisition time window, detect the spike trigger frequency of the target feature, analyze the fluctuation of the feature correlation for visual target classification, identify the feature change trend, filter out the continuously changing features as stable features, and generate the feature change classification result; S2: Invoke the feature change classification result, analyze the trans-layer connection paths of neurons for the stable features, adjust the synaptic transmission ability, avoid signal transmission delay, filter out the synaptic connection paths of the unstable features, reduce the priority of the trans-layer connections, and generate the trans-layer connection adjustment result; S3: Invoke the trans-layer connection adjustment result, count the time distribution of the spike events, refer to the spike time pattern of the input sample, analyze the degree of offset of the spike time, for the dynamic target recognition scenario, adjust the time reference point, evaluate the impact of the time offset on the spike time distribution, and obtain the time reference point adjustment result; S4: Adopt the time reference point adjustment result, calculate the time scaling ratio according to the spike trigger time of the input sample, adjust the spike time of the input data according to the time scaling ratio, make the spike events match the same time scale, and generate the time-aligned spike data.

[0006] As a further solution of the present invention, the feature change classification result includes stable feature records, unstable feature records, and feature fluctuation analysis results. The trans-layer connection adjustment result includes synaptic transmission adjustment information, priority adjustment results of the connection paths, and signal delay increments. The time reference point adjustment result includes time distribution analysis results, time offset evaluation information, and dynamic adjustment conditions. The time-aligned spike data includes interval consistency, scaling ratio implementation conditions, and time scale matching information.

[0007] As a further solution of the present invention, the specific steps for obtaining the feature change classification result are as follows: S101: Collect the spike neuron firing data within the acquisition time window, detect the spike trigger frequency of the target feature, extract the spike train data, identify the trigger frequency distribution and changes of the target feature, and obtain the target feature spike response record; S102: Based on the target feature spike response record, analyze the fluctuation amplitude and change trend of the target feature, classify the feature change types, filter out the target features with stable change trends and fluctuation amplitudes within the stable range as stable features, and record the target features with unstable change trends and outside the stable range, to obtain the target feature change trend recognition result; S103: Invoke the target feature change trend recognition result, filter out the continuously changing target features, record the fluctuation trend and frequency changes, analyze the fluctuation amplitude of the spike trigger frequency, and generate the feature change classification result.

[0008] As a further solution of the present invention, the step of obtaining the cross-layer connection adjustment result is specifically as follows: S201: Based on the feature change classification result, analyze the connection paths of stable features between different neuron layers, determine whether the synaptic signals match the signal transmission requirements, optimize the synaptic transmission ability and increase the path signal transmission time to obtain the synaptic signal matching result; S202: Invoke the synaptic signal matching result, screen the synaptic connection paths of non-stable features, analyze the connection priority of cross-layer transmission, adjust the signal transmission time, weaken the synaptic connections within the signal adjustment range, screen the cross-layer connection paths that meet the requirements after optimization, and generate the cross-layer connection adjustment result.

[0009] As a further solution of the present invention, the step of obtaining the time reference point adjustment result is specifically as follows: S301: Invoke the cross-layer connection adjustment result, count the time distribution of pulse events, obtain the time point sequence of multi-pulse events, and analyze the distribution of pulse events in the time dimension with reference to the pulse time pattern of the input sample to obtain the pulse time distribution density; S302: Based on the pulse time distribution density, analyze the time span range and the pulse time offset degree, evaluate the time alignment degree between the pulse time offset degree and the time pattern of the input sample, adjust the time reference point for the dynamic target recognition scenario, match the target motion trajectory, and compare the change amplitude of the time alignment error before and after adjustment to obtain the time reference point error record; S303: Invoke the time reference point error record, evaluate the impact of the time offset on the pulse time distribution according to the time offset degree of the target motion trajectory, and adjust the time reference point to generate the time reference point adjustment result.

[0010] As a further solution of the present invention, the evaluation of the time alignment degree between the pulse time offset degree and the time pattern of the input sample adopts the formula: ; where D align represents the time alignment error, T pulse,i represents the i-th pulse time value, T ref,i represents the i-th reference time value, W i represents the pulse weight at the i-th time point, and N represents the total number of pulse samples.

[0011] As a further solution of the present invention, the step of obtaining the time-aligned pulse data is specifically as follows: S401: Invoke the time reference point adjustment result, calculate the time interval of pulse events, extract the pulse trigger time of the input sample, and calculate the mean value of the pulse trigger time interval to obtain the average time interval of pulse events; S402: Calculate a time scaling ratio based on the average time interval between pulse events, adjust the pulse trigger times of the input samples according to the time scaling ratio to make the pulse events match the same time scale, evaluate the distribution consistency of the adjusted pulse times, verify the uniformity of the pulse sequence on the time axis, and obtain time-aligned pulse data.

[0012] As a further solution of the present invention, the formula for adjusting the pulse trigger time of the input sample according to the time scaling ratio is: ; where t' i represents the adjusted trigger time of the i-th pulse, t i represents the original trigger time of the i-th pulse, t min represents the minimum value among the pulse trigger times, t max represents the maximum value among the pulse trigger times, and τ represents the total duration of the target time scale.

[0013] As a further solution of the present invention, the method further includes step S5: S5: Based on the time-aligned pulse data, evaluate the time stability of multi-category pulse events, statistically analyze the distribution deviation of multi-category pulse events within a time window, determine the matching priority of pulse events, calculate the time stability score of the visual target category, adjust the matching degree according to the score, generate a time-adapted classification result, optimize the adaptability of the pulse neural network to dynamic visual scenes, and perform target recognition and classification; The time-adapted classification result includes the time stability score of the pulse event, the adjusted result of the matching priority, and the statistical information of the distribution deviation.

[0014] As a further solution of the present invention, the specific steps for obtaining the time-adapted classification result are as follows: S501: Based on the time-aligned pulse data, statistically analyze the distribution of multi-category pulse events within a time window, analyze the time distribution characteristics of different-category pulse events, and obtain the time deviation value of the pulse event; S502: Invoke the time deviation value of the multi-category pulse event, analyze the matching priority of the multi-category pulse event, calculate the time stability score of the visual target category according to the matching priority, analyze the influence of the time stability score on the pulse event matching, and adjust the matching degree to obtain a matching time correction result; S503: Based on the matching time correction result, adjust the matching method of the pulse event, generate a time-adapted classification result according to the adjusted time distribution of the pulse event, optimize the adaptability of the pulse neural network to dynamic visual scenes, and perform target recognition and classification.

[0015] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least: By collecting and analyzing the excitation data of spiking neurons, data showing continuous changes or unstable features within the time window are screened out, effectively improving the dynamic adaptability of feature recognition. Adjust the cross-layer connection paths and synaptic transmission capabilities of neurons, optimize for stable and unstable features, and further improve the efficiency and accuracy of visual target classification, especially in dynamic environments. Based on statistical analysis and adjustment of pulse time, the recognition ability of spiking neural networks for time-sensitive targets is enhanced, so that time-aligned pulse data can more accurately reflect the real-time changes of visual input. Through this refined time and feature processing, the target recognition and classification effects in dynamic scenes are significantly optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the workflow of the present invention; DETAILED DESCRIPTION

[0017] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0018] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0019] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0020] See also Figure 1 The embodiment of the present invention provides a neuromorphic visual target classification method based on a spiking neural network. The processing flow of the method may include the following steps: S1: Collect the pulse neuron excitation data within the time window, detect the pulse trigger frequency of the target feature, analyze the feature correlation fluctuation of the visual target classification, identify the feature change trend, screen the continuously changing features as stable features, record the features with uneven change amplitude as unstable features, and generate the feature change classification results; S2: Call the classification results of feature changes, analyze the cross-layer connection paths of neurons with stable features, adjust the synaptic transmission capacity, optimize the pulse propagation efficiency of visual target classification, avoid signal transmission delay, screen the synaptic connection paths of unstable features, reduce the cross-layer connection priority, increase the path signal delay, weaken the synaptic plasticity, and generate the cross-layer connection adjustment results; S3: Invoke the cross-layer connection adjustment result, count the time distribution of pulse events, refer to the pulse time pattern of the input sample, analyze the time span range and the offset degree of the pulse time, for the dynamic target recognition scenario, adjust the time reference point, refer to the time offset degree of the target motion trajectory, evaluate the impact of the time offset on the pulse time distribution, and obtain the time reference point adjustment result; S4: Adopt the time reference point adjustment result to identify the average interval of pulse events, calculate the time scaling ratio according to the pulse trigger time of the input sample, and adjust the pulse time of the input data according to the time scaling ratio to make the pulse events match the same time scale, and generate time-aligned pulse data; S5: Based on the time-aligned pulse data, evaluate the time stability of multi-category pulse events, count the distribution deviation of multi-category pulse events within the time window, judge the matching priority of pulse events, calculate the time stability score of the visual target category, adjust the matching degree according to the score, generate the time-adapted classification result, optimize the adaptation ability of the pulse neural network to the dynamic visual scene, and perform target recognition and classification; The feature change classification result includes stable feature records, unstable feature records, and feature fluctuation analysis results. The cross-layer connection adjustment result includes synaptic transmission adjustment information, connection path priority adjustment results, and signal delay increments. The time reference point adjustment result includes time distribution analysis results, time offset evaluation information, and dynamic adjustment conditions. The time-aligned pulse data includes interval consistency, scaling ratio implementation, and time scale matching information. The time-adapted classification result includes the time stability score of pulse events, matching priority adjustment results, and distribution deviation statistical information.

[0021] The steps for obtaining the feature change classification result are specifically as follows: S101: Collect the pulse neuron excitation data within the time window, detect the pulse trigger frequency of the target feature, extract the pulse sequence data, identify the trigger frequency distribution and change of the target feature, and obtain the target feature pulse response record; It is necessary to continuously sample the neuron excitation signal, set the sampling frequency to 10 kHz to ensure that all high-frequency pulse signals can be captured. After signal acquisition, use the band-pass filtering method to remove low-frequency noise and high-frequency interference. After filtering, perform pulse detection on the data. The pulse trigger frequency is calculated using the counting method, that is, counting the number of pulses triggered by neurons within the unit time window (100 ms). The calculation formula is: ; where F t represents the pulse trigger frequency within the unit time window, and N t is the number of pulses detected within the time window T; If 50 pulses are detected within 100 ms, then: ; It is necessary to extract the pulse sequence data, store the timestamp and pulse amplitude information, use the time series analysis method to statistically analyze the trigger frequency distribution of the target feature, and obtain parameters such as its mean, variance, and kurtosis. The mean is used to measure the central tendency of pulse triggering, and the variance is used to measure the frequency fluctuation. The threshold setting basis: In the study of biological neuron excitation, a variance less than 50 is set as a stable signal, and a variance exceeding 200 is set as a large fluctuation signal. This threshold is statistically based on experimental data and is obtained by calculating the standard deviation of 50 records. If the measured variance is greater than this value, it is necessary to further analyze its change trend. The pulse trigger frequency of a certain neuron target feature fluctuates around 500, and the measured standard deviation is 8, which is lower than 50, then it can be determined that it is a stable signal. If the standard deviation of a certain feature reaches 220, it is determined that the signal has large fluctuations, and a target feature pulse response record is formed.

[0022] S102: Based on the target feature pulse response record, analyze the fluctuation amplitude and change trend of the target feature, classify the feature change types, and screen out the target features with stable change trends and fluctuation amplitudes within the stable range and classify them as stable features. Record the target features with unstable change trends and outside the stable range to obtain the target feature change trend recognition result; Analyze the fluctuation amplitude and change trend of the target feature, calculate the fluctuation amplitude, that is, the difference between the maximum and minimum pulse frequencies, and the calculation formula is: ; Among them, the parameter F max is the maximum pulse frequency within the time window, and F min is the minimum pulse frequency; If the maximum value of the pulse frequency of a certain target feature within 100 ms is 800 Hz and the minimum value is 500 Hz, then its fluctuation amplitude is: ; Identify the change trend, use the moving average method for trend analysis, calculate the average value of the first five windows, and so on; Stable interval threshold setting: In bioelectrical signal processing, the stable interval is set within ±10% of the mean of the target feature. That is, if the mean of the target feature is 600 Hz, the stable interval is set from 540 Hz to 660 Hz. If the fluctuation exceeds this range, it is judged as an unstable feature. If the frequency mean of a certain target feature is 550 Hz and the measured data are [510, 520, 530, 515, 525] Hz respectively, then all the data are within the stable interval (495 Hz to 605 Hz), so this feature can be determined as a stable feature. However, if the measured data are [510, 650, 490, 700, 450] Hz, which exceeds the stable interval range, further processing is required to obtain the recognition result of the target feature change trend.

[0023] S103: Invoke the recognition result of the target feature change trend, screen out the target features with continuous changes, record the fluctuation trend and frequency change, analyze the fluctuation amplitude of the pulse trigger frequency, and generate the feature change classification result; Screen out the target features with continuous changes. Traverse the pulse trigger frequency sequences of all target features, extract the features that continuously change within multiple time windows, record the frequency data of the features, identify the fluctuation trend of each feature, determine the overall direction of frequency increase or decrease, analyze the amplitude of frequency change, observe the change range by calculating the differences between each time point, and evaluate the change rate. If the fluctuation range of a certain target feature is large and it fails to stabilize within a long time, further analyze its frequency change pattern. If the trigger frequency of a certain feature shows an increasing trend in consecutive multiple time windows and its increase amplitude continues to expand, it can be judged that this feature has a significant growth trend. If a certain feature shows frequent up and down fluctuations in different time windows, it is necessary to further judge whether there is a periodic change. By comparing the change situations of multiple target features, generate the feature change classification result.

[0024] The specific steps for obtaining the cross-layer connection adjustment result are as follows: S201: Based on the feature change classification result, analyze the connection paths of stable features between different neuron layers, judge whether the synaptic signals match the signal transmission requirements, optimize the synaptic transmission ability and increase the path signal transmission time to obtain the synaptic signal matching result; Analyze the connection paths of stable features between different neuron layers, construct the neuron network topology structure, determine the neurons to which the stable features belong and their synaptic connection situations. The connection path of each neuron node is represented by an adjacency matrix A. Let A(i,j)=1 indicate that there is a synaptic connection between neuron i and j, and A(i,j)=0 indicate no connection. Further calculate the synaptic signal transmission delay. Assume that the synaptic delay time from neuron i to j is T(i,j), and the calculation formula is: ; where d(i, j) is the synaptic distance between neurons i and j, and V s is the signal transmission speed; If d(i, j) = 0.2 mm and V s = 1.5 m / s, then: ; Determine whether the synaptic signal matches the signal transmission requirement, and calculate the matching error Em of the synaptic signal. Setting of synaptic signal matching error: Set the error range of synaptic delay time within ±10% of the average transmission time of the target neuron. That is, if the average transmission time of the neuron is 1.2 ms, the synaptic signal matching error is set between 1.08 ms and 1.32 ms. If the error exceeds this range, it is considered that the synaptic signal does not match. If the average transmission time of a certain neuron is 1.5 ms, but the calculated synaptic delay time is 2.0 ms, the error is 0.5 ms, which exceeds the matching error range. Then, it is necessary to optimize the synaptic transmission ability. The optimization methods include adjusting the synaptic weighting coefficient to improve the signal transmission efficiency, or adding new connection paths to reduce the transmission time, to obtain the synaptic signal matching result.

[0025] S202: Invoke the synaptic signal matching result, screen the synaptic connection paths with non - stable features, analyze the connection priority of cross - layer transmission, adjust the signal transmission time, weaken the synaptic connections within the signal adjustment range, screen the cross - layer connection paths that meet the requirements after optimization, and generate the cross - layer connection adjustment result; Screen the synaptic connection paths with non - stable features, extract all neurons with non - stable features, and determine their cross - layer transmission paths according to topological structure A. Calculate the cross - layer transmission priority P(i, j), and set the calculation formula as: ; where W syn (i, j) is the synaptic weight from neuron i to j, and T(i, j) is the synaptic signal transmission time; The higher the priority, the more important this connection path is in cross - layer transmission; If W syn (i, j) = 0.8 and T(i, j) = 1.2 ms, then P(i, j) = 0.67; If W syn (i, j) = 0.6 and T(i, k) = 1.8 ms, then P(i, k) = 0.33; Therefore, the priority of connection (i, j) is higher than that of connection (i, k); Priority Threshold Setting: According to the adaptive mechanism of the biological neural network, the threshold for setting the cross-layer transmission priority is set to 0.5. If the priority of the synaptic connection path is lower than this value, the signal transmission time is adjusted, and its influence is reduced by weakening the synaptic connection weight. If the priority of a certain synaptic connection path is 0.3, its synaptic weight is weakened to 0.1 to reduce its influence. Finally, the cross-layer connection paths that meet the requirements after screening and optimization are selected to generate the cross-layer connection adjustment result.

[0026] The steps for obtaining the time reference point adjustment result are specifically as follows: S301: Invoke the cross-layer connection adjustment result, count the time distribution of pulse events, obtain the time point sequence of multi-pulse events, analyze the distribution of pulse events in the time dimension with reference to the pulse time pattern of the input sample, and obtain the pulse time distribution density; Perform time statistics on pulse events, extract the occurrence time of pulse events, and form a time series. For this time series, set a time window. For example, count the number of pulse events within a 100-millisecond time window, calculate the density of pulse events per unit time. For the distribution of pulse events, use a histogram to count the number of pulses in different time periods, and use the method of Gaussian distribution fitting for distribution modeling to form the time distribution curve of pulse events. Within a certain time window, if 50 pulse events are counted, and the average occurrence frequency of the overall pulse events is 40 times per 100 milliseconds, then the pulse density within this window is higher than the overall average value. It can be judged that some time periods are high-incidence areas of pulse events. At the same time, with reference to the pulse time pattern of the input sample, compare and analyze the sample pulse sequence with the statistically obtained pulse distribution density, set the time step, slide-align the sample pulse time pattern with the statistical distribution, and calculate the matching degree of pulse events at each time step. If an input sample contains 40 pulse events in the interval from 0 to 100 milliseconds, and the number of pulse events in the same time interval in the statistical data is 45 times, then the matching degree is calculated as 40 divided by 45, and the result is approximately 0.89, so as to obtain the matching degree between the sample pattern and the statistical data. In this process, to ensure that the statistical density of pulse events meets the standards of the target recognition device, a pulse density threshold needs to be set. This threshold is calculated based on the requirements of the target device for the pulse event frequency. If the target recognition device requires a minimum resolvable time interval of 5 milliseconds, then this device needs at least 200 pulse events per second. Therefore, the pulse density threshold is set to 200 times per second. If the pulse density within a certain time window is lower than this threshold, it is regarded as a pulse sparse area and does not participate in the pattern matching calculation. If the minimum sampling interval of the device is set to 10 microseconds, the setting of the density threshold can be further refined to 100,000 times per second to ensure that the calculation result is within the range that the device can handle, and adjust it in combination with the matching situation between the input sample and the pulse time distribution density to obtain the pulse time distribution density.

[0027] S302: Analyze the time span range and the degree of pulse time offset based on the pulse time distribution density, evaluate the time alignment degree between the degree of pulse time offset and the time pattern of the input sample, adjust the time reference point for the dynamic target recognition scenario, match the target motion trajectory, and compare the change range of the time alignment error before and after adjustment to obtain the time reference point error record; Evaluate the time alignment degree between the degree of pulse time offset and the time pattern of the input sample, using the formula: ; where D align represents the time alignment error, T pulse,i represents the time value of the i-th pulse, T ref,i represents the i-th reference time value, W i represents the pulse weight at the i-th time point, and N represents the total number of pulse samples; Parameter explanation and calculation process: Pulse time value T pulse,i : Record the time point when the i-th pulse arrives through a high-precision time measurement device. In one measurement, the recorded pulse time value is: ; Reference time value T ref,i : Set the reference time value of the i-th pulse according to the designed or expected pulse arrival time. Corresponding to the above pulse time, the reference time value is: ; Pulse weight W i : Set the weight according to the importance or signal strength of each pulse. The basis for setting the weight can be factors such as signal strength and signal-to-noise ratio. Through measurement, the obtained weight value is: ; Calculate the absolute value of the weighted time deviation of each pulse: For the first pulse: ; For the second pulse: ; ; For the third pulse: ; Calculate the sum of the absolute values of the weighted time deviations: ; Calculate the square root of the sum of the squares of the weights: ; Calculate the time alignment error value: ; The result shows that after calculation, the time alignment error value is approximately 0.000639 seconds. This value reflects the weighted average deviation degree between the pulse time and the reference time. The smaller the value, the higher the alignment degree between the pulse time and the reference time.

[0028] S303: Call the time reference point error record, evaluate the impact of time offset on the pulse time distribution according to the time offset degree of the target motion trajectory, adjust the time reference point, and generate the time reference point adjustment result; According to the time offset degree of the target motion trajectory, compare the pulse time distributions before and after adjustment, calculate the mean values of the pulse time distributions before and after adjustment. If the mean value before adjustment is 58 milliseconds and the mean value after adjustment is 62 milliseconds, calculate the offset of the pulse time distribution, which is equal to the mean value after adjustment minus the mean value before adjustment, getting 4 milliseconds. Evaluate the impact of time offset on the pulse time distribution, calculate the mean square errors before and after adjustment. For example, if the mean square error before adjustment is 10 millisecond squares and the mean square error after adjustment is 6 millisecond squares, then calculate the error reduction amount as 4 millisecond squares. In this process, to ensure the rationality of the adjustment, it is necessary to set the mean square error threshold. The setting of this threshold is based on the acceptable range of time alignment error of the device. If the maximum acceptable alignment error of the device is 5 millisecond squares, then set the error threshold to 5 millisecond squares. If the calculated error reduction value is still higher than this threshold, then it is necessary to further optimize the time reference point. If the mean square error after adjustment is lower than this threshold, if the calculated result is 3 millisecond squares, then it is considered that the time alignment has reached the acceptable range and there is no need to further adjust the time reference point, and generate the time reference point adjustment result.

[0029] The specific steps for obtaining time-aligned pulse data are as follows: S401: Call the time reference point adjustment result, calculate the time interval of pulse events, extract the pulse trigger time of the input sample, and calculate the mean value of the pulse trigger time intervals to obtain the average pulse event time interval; Calculate the time intervals between pulse events, extract the pulse trigger times of the input samples, arrange them in chronological order, and calculate the time intervals between adjacent pulse events. If the pulse trigger times of a certain input sample are 10 milliseconds, 25 milliseconds, 45 milliseconds, and 70 milliseconds in sequence, the adjacent time intervals are 15 milliseconds, 20 milliseconds, and 25 milliseconds respectively. Calculate the mean of the pulse trigger time intervals, which is defined as the sum of all time intervals divided by the number of intervals. The average time interval of the above data is calculated as 60 milliseconds divided by 3, and the result is 20 milliseconds. During this process, to ensure that the calculation of the pulse event intervals meets the time resolution requirements of the device, a time interval reference value needs to be set. This reference value is set based on the minimum sampling period or pulse trigger mechanism of the device. In a certain neuromorphic computing device, the time resolution is set to 1 millisecond, so the time interval reference value should be set as an integer multiple of 1 millisecond, such as 10 milliseconds, 20 milliseconds, etc. If the calculated average time interval of the pulse events is 19 milliseconds, the device will automatically adjust to the closest reference value of 20 milliseconds to meet the device timing requirements. This reference value setting can be optimized according to the device operating environment. In a high-load environment, the time interval reference value needs to be increased to 50 milliseconds to reduce the computational overhead of pulse events and obtain the average time interval of pulse events.

[0030] S402: Based on the average time interval of pulse events, calculate the time scaling ratio, adjust the pulse trigger times of the input samples according to the time scaling ratio, make the pulse events match the same time scale, evaluate the distribution consistency of the adjusted pulse times, verify the uniformity of the pulse sequence on the time axis, and obtain the time-aligned pulse data; The formula for adjusting the pulse trigger time of the input sample according to the time scaling ratio is: ; where, t' i represents the adjusted trigger time of the i-th pulse, t i represents the original trigger time of the i-th pulse, t min represents the minimum value among the pulse trigger times, t max represents the maximum value among the pulse trigger times, and τ represents the total duration of the target time scale; Parameter meanings and calculation processes: The parameter t i represents the original trigger time of the i-th pulse and is directly obtained from the pulse event data monitoring device; t min and t max are respectively the minimum and maximum values among all pulse trigger times, which are calculated by counting the time points in the pulse data set; τ is the total duration of the target time scale, which is determined according to the experimental design requirements and reflects the length of the time frame to which it needs to be adjusted; Taking actual data as an example, set the time data for obtaining a set of pulse events from a sensor network, and monitor the following time points (unit: millisecond): {100, 150, 200, 250, 300}; For this set of data, calculate t min = 100 and t max = 300, and set the target time scale τ = 1 second (1000 milliseconds); Calculate the adjusted time of the first pulse: ; Calculate the adjusted time of the second pulse: ; Calculate the adjusted time of the third pulse: ; Calculate the adjusted time of the fourth pulse: ; Calculate the adjusted time of the fifth pulse: ; The above calculation results show how to adjust the original pulse trigger time to different time points within 1 second. The results indicate that the adjusted pulse times can be evenly distributed on the set time scale, meeting the requirement of matching pulse events to the same time scale, which helps to further analyze the distribution consistency of pulse times and the uniformity of sequences. The adjusted time points will be used to evaluate the consistency of pulse time distribution and verify the uniformity of pulse data after time alignment.

[0031] The specific steps for obtaining the time adaptation classification results are as follows: S501: Based on the time-aligned pulse data, statistically analyze the distribution of multi-category pulse events within the time window, analyze the time distribution characteristics of different-category pulse events, and obtain the time deviation values of multi-category pulse events; For the statistics of different categories of pulse events within a time window, a fixed time window needs to be set, set to 100 milliseconds, and all pulse events are segmented according to this window to ensure that all events can be statistically analyzed on the same time scale. Within each time window, the number of pulse events of different categories is counted to form time series data. Suppose in the first time window, there are 15 pulse events of category A, 8 of category B, and 20 of category C, then the data record for this window is [15, 8, 20]. After accumulating the data of all windows, a time distribution matrix is formed. It is necessary to calculate the time deviation values of different categories of pulse events. The time deviation value can be obtained by calculating the fluctuation of this category within all time windows. If the number of pulse events of category A in 10 time windows is 10, 12, 8, 15, 11, 14, 9, 13, 10, 12 in sequence, then its overall fluctuation can be calculated to obtain the time deviation value of this category. The time deviation values of the remaining categories can be calculated in the same way and normalized to make the time deviation values of all categories within the same magnitude range. After the calculation is completed, the time deviation value of the pulse event is obtained.

[0032] S502: Call the time deviation values of multi-category pulse events, analyze the matching priorities of multi-category pulse events, calculate the time stability score of the visual target category according to the matching priorities, analyze the impact of the time stability score on pulse event matching, adjust the matching degree, and obtain the matching time correction result; Analyze the matching priorities of pulse events of various categories, determine the priority weights based on the time deviation values. The setting of the weights can adopt an inverse relationship, that is, the smaller the time deviation value of a category, the greater its priority weight. Suppose the normalized time deviation value of a certain category is 0.2, then its matching priority weight is high, while the time deviation value of another category is 0.5, then its matching priority is low. Calculate the time stability score of the visual target category. This score is related to the time deviation value and the priority weight. If a certain category has a high priority weight but still has a large time deviation value, then the time stability score of this category will be affected to a certain extent. After the calculation of the time stability score is completed, an impact analysis needs to be carried out. Set an impact threshold, set to 0.6. If the score of a certain category is lower than this value, then reduce its matching degree, adjust the matching method, weaken the matching degree by a certain proportion, so that the matching priority of this category decreases. After the adjustment is completed, obtain the matching time correction result, and reassign the category matching priorities according to the score after the matching adjustment to obtain the matching time correction result.

[0033] S503: Based on the matching time correction result, adjust the matching method of pulse events, generate a time adaptation classification result according to the adjusted time distribution of pulse events, optimize the adaptation ability of the pulse neural network to dynamic visual scenes, and perform target recognition and classification; Adjust the matching method of pulse events, establish a time distribution adaptation matrix, which is used to redistribute the time matching ratio of pulse events. If the matching score of a certain category is high, then in the final time adaptation classification, the number of pulse events in this category will increase within a certain range. If the matching scores of categories A, B, and C are 0.7, 0.5, and 0.9 respectively, then the time distribution adaptation result will tend to the category with a high matching score. Adjust the pulse event matching method according to this matrix. If the original event quantity matrix is [15, 8, 20], it will become [13, 9, 21] after adjustment, making the event distribution more in line with the matching priority requirements. Generate a time adaptation classification result according to the adjusted pulse event time distribution, and optimize the time adaptation parameters of the pulse neural network. The optimization method can adjust the matching degree through normalization to make the matching coefficients of each category reach a reasonable range. If the matching coefficients are 0.8, 0.75, and 0.9 respectively, they will be adjusted to [0.84, 0.79, 0.91] after optimization, completing the matching adjustment and classification optimization of pulse events.

[0034] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A neuromorphic visual target classification method based on a spiking neural network, characterized in that: The following steps are involved: S1: Collect pulse neuron excitation data within the time window, detect the pulse trigger frequency of the target feature, analyze the feature correlation fluctuation of visual target classification, identify the feature change trend, screen the continuously changing features as stable features, and generate feature change classification results; S2: calling the feature change classification result, analyzing the cross-layer connection path of neurons with stable features, adjusting the synaptic transmission capacity, avoiding signal transmission delay, screening the synaptic connection path of unstable features, reducing the cross-layer connection priority, and generating a cross-layer connection adjustment result; S3: calling the cross-layer connection adjustment result, counting the time distribution of the pulse event, referring to the pulse time mode of the input sample, analyzing the offset degree of the pulse time, adjusting the time reference point for the dynamic target recognition scenario, evaluating the influence of the time offset on the pulse time distribution, and obtaining the time reference point adjustment result; S4: Using the time reference point adjustment result, according to the pulse trigger time of the input sample, calculate the time scaling ratio, adjust the input data pulse time according to the time scaling ratio, so that the pulse events match the same time scale, and generate time-aligned pulse data.

2. The neuromorphic visual target classification method based on spiking neural network according to claim 1 is characterized in that: The feature change classification results include stable feature records, unstable feature records, and feature fluctuation analysis results; the cross-layer connection adjustment results include synaptic transmission adjustment information, connection path priority adjustment results, and signal delay increments; the time reference point adjustment results include time distribution analysis results, time offset evaluation information, and dynamic adjustment status; the time-aligned pulse data includes interval consistency, scaling implementation status, and time scale matching information.

3. The neuromorphic visual target classification method based on spiking neural network according to claim 1 is characterized in that: The steps for obtaining the feature change classification result are specifically as follows: S101: Collecting pulse neuron excitation data within a time window, detecting the pulse trigger frequency of the target feature, extracting pulse sequence data, identifying the distribution and change of the trigger frequency of the target feature, and obtaining the target feature pulse response record; S102: Based on the target feature pulse response record, analyzing the fluctuation amplitude and change trend of the target feature, classifying the feature change type, screening target features with stable change trend and fluctuation amplitude in a stable range as stable features, recording target features with unstable change trend and beyond the stable range, and obtaining target feature change trend recognition results; S103: calling the target feature change trend recognition result, screening the continuously changing target features, recording the fluctuation trend and frequency change, analyzing the fluctuation amplitude of the pulse trigger frequency, and generating the feature change classification result.

4. The neuromorphic visual target classification method based on spiking neural network according to claim 1 is characterized in that: The steps of obtaining the cross-layer connection adjustment result are specifically as follows: S201: Based on the feature change classification results, analyze the connection paths of stable features between differentiated neuron layers, determine whether the synaptic signal matches the signal transmission requirements, optimize the synaptic transmission capacity and increase the path signal transmission time, and obtain the synaptic signal matching results; S202: calling the synaptic signal matching result, screening the synaptic connection path with unstable characteristics, analyzing the connection priority of cross-layer transmission, adjusting the signal transmission time, weakening the synaptic connection within the signal adjustment range, screening the cross-layer connection path that meets the requirements after optimization, and generating a cross-layer connection adjustment result.

5. The neuromorphic visual target classification method based on spiking neural network according to claim 1, characterized in that: The steps for obtaining the time reference point adjustment result are specifically as follows: S301: calling the cross-layer connection adjustment result, counting the time distribution of pulse events, obtaining the time point sequence of multiple pulse events, referring to the pulse time mode of the input sample, analyzing the distribution of the pulse events in the time dimension, and obtaining the pulse time distribution density; S302: Based on the pulse time distribution density, analyze the time span range and the pulse time deviation degree, evaluate the pulse time deviation degree and the time alignment degree of the input sample time pattern, adjust the time reference point for the dynamic target recognition scenario, match the target motion trajectory, and compare the change amplitude of the time alignment error before and after the adjustment to obtain the time reference point error record; S303: calling the time reference point error record, evaluating the influence of the time offset on the pulse time distribution according to the time offset degree of the target motion trajectory, adjusting the time reference point, and generating a time reference point adjustment result.

6. The neuromorphic visual target classification method based on spiking neural network according to claim 5 is characterized in that: The evaluation of the degree of pulse time deviation and the degree of time alignment of the input sample time pattern is performed using the formula: ； Among them, D align represents the time alignment error, T pulse,i represents the time value of the ith pulse, T ref,i represents the i-th reference time value, W i represents the pulse weight at the i-th time point, and N represents the total number of pulse samples.

7. The neuromorphic visual target classification method based on spiking neural network according to claim 1, characterized in that: The steps for acquiring the time-aligned pulse data are specifically as follows: S401: calling the time reference point adjustment result, calculating the time interval of the pulse event, extracting the pulse trigger time of the input sample, calculating the mean of the pulse trigger time interval, and obtaining the average time interval of the pulse event; S402: Based on the average time interval of the pulse events, calculate the time scaling ratio, adjust the pulse trigger time of the input sample according to the time scaling ratio, so that the pulse events match the same time scale, evaluate the distribution consistency of the adjusted pulse time, verify the uniformity of the pulse sequence on the time axis, and obtain time-aligned pulse data.

8. The neuromorphic visual target classification method based on spiking neural network according to claim 7 is characterized in that: The formula for adjusting the pulse trigger time of the input sample according to the time scaling ratio is: ； Among them, t' i represents the adjusted trigger time of the ith pulse, t i represents the original trigger time of the ith pulse, t min Represents the minimum value of the pulse trigger time, t max represents the maximum value of the pulse trigger time, and τ represents the total duration of the target time scale.

9. The neuromorphic visual target classification method based on spiking neural network according to claim 1, characterized in that: The method further comprises step S5: S5: Based on the time-aligned pulse data, evaluate the temporal stability of multi-category pulse events, count the distribution deviations of multi-category pulse events in the time window, determine the pulse event matching priority, calculate the temporal stability score of the visual target category, adjust the matching degree according to the score, generate a temporal adaptation classification result, optimize the adaptability of the pulse neural network to dynamic visual scenes, and perform target recognition and classification; The time adaptation classification result includes the time stability score of the pulse event, the matching priority adjustment result, and the distribution deviation statistical information.

10. The neuromorphic visual target classification method based on spiking neural network according to claim 9, characterized in that: The steps for obtaining the time adaptation classification result are specifically as follows: S501: Based on the time-aligned pulse data, count the distribution of multiple categories of pulse events in the time window, analyze the time distribution characteristics of differentiated category pulse events, and obtain time deviation values ​​of multiple categories of pulse events; S502: calling the time deviation values ​​of the multi-category pulse events, analyzing the matching priorities of the multi-category pulse events, calculating the time stability scores of the visual target categories according to the matching priorities, analyzing the influence of the time stability scores on the matching of the pulse events, adjusting the matching degree, and obtaining the matching time correction result; S503: Based on the matching time correction result, the matching method of the pulse event is adjusted, and according to the adjusted pulse event time distribution, a time adaptation classification result is generated to optimize the adaptability of the pulse neural network to dynamic visual scenes, and perform target recognition and classification.

Images