Intelligent shooting behavior prediction and optimization method based on deep learning

Abstract

The present invention relates to the field of behavior prediction technology, specifically to an intelligent shooting behavior prediction and optimization method based on deep learning, comprising the following steps: acquiring stretching data to analyze displacement trends, identifying synchronization of same-direction enhancement and mutation, collecting aiming speed, grip strength, recoil and breathing data to analyze collaborative conflicts, extracting angle changes to classify aiming trends, integrating three types of labels to judge behavioral bias, matching prompt content and outputting prediction results. In the present invention, by identifying changes in the stretching trend of the perception area, enhancing the perception of the structural change state, integrating the dynamic sequence of aiming speed, grip strength, recoil response and breathing rhythm, mining the synergistic relationship of behavioral driving factors, extracting the frequency of aiming angle deviation, identifying behavioral trend characteristics, horizontally comparing multiple types of labels, clarifying the direction of behavioral bias, matching trend attribution to prompt content and then classifying and outputting, the full process closed loop from multi-source data analysis to prediction guidance improves the accuracy, real-time performance and tactical adaptability of the prediction.

Description

Technical Field

[0001] The present invention relates to the field of behavior prediction technology, and in particular to an intelligent shooting behavior prediction and optimization method based on deep learning. Background Art

[0002] The field of behavior prediction technology encompasses methods and techniques for inferring the future behavior of individuals or groups based on data analysis and pattern recognition. The core of this technology lies in the use of artificial intelligence algorithms such as machine learning and deep learning to extract features from historical behavioral data and, through modeling, achieve probabilistic predictions of future behavior. Behavior prediction technology is primarily used in scenarios such as intelligent security, traffic forecasting, human-computer interaction, military simulation, and intelligent decision-making. Through the continuous learning of multi-dimensional dynamic data, this field achieves accurate modeling of behavioral patterns and, combined with sensory data, time series analysis, and probabilistic inference, predicts changing trends in target behavior. This field is characterized by data-driven, model-adaptive, and real-time inference. Its development relies on the continuous improvement of large-scale data acquisition capabilities, computing resource support, and algorithm optimization capabilities.

[0003] Among them, the intelligent shooting behavior prediction and optimization method based on deep learning refers to the use of deep neural network structures to extract features from multi-source dynamic data during the shooting process, and to predict the shooting action's spatial trajectory, posture changes, firing intention and other behavioral parameters by constructing a time series association model. At the same time, it combines historical shooting data to model target recognition, target tracking, and the evolution trend of shooting intention. This method specifically relies on convolutional neural networks to extract visual information, processes time series data through recurrent neural networks, and jointly utilizes reinforcement learning algorithms to continuously optimize shooting strategies. In addition, this method also forms a high-dimensional action prediction model based on large-scale shooting behavior data training, which is used to infer the state transition of individual behaviors during the shooting process and to perform real-time corrections on the prediction results.

[0004] Current technical approaches mostly focus on probabilistic modeling based on static behavioral data, lacking the ability to fine-tune the temporal evolution of microstructural states and making it difficult to effectively capture the interconnected changes between multiple regions within a launcher. Conventional approaches often use single variables or low-dimensional combinations to predict firing behavior, ignoring the co-evolution of multiple behavioral variables in the temporal dimension. This leads to logical discontinuities in analyzing the causes of complex behavior. The frequency of directional shifts in aiming behavior is not systematically modeled, limiting in-depth identification of fluctuation patterns in the behavior and resulting in trend misjudgment and delayed stability assessments. Prediction results are mostly static outputs lacking a structured linkage between behavioral prompts, making it difficult to integrate the output information into operational decisions and easily leading to inefficient communication between identified content and action commands. In a tactical environment requiring high-speed feedback and real-time response, this type of technical approach struggles to ensure the continuity of prediction logic and the operational value of behavioral identifiers, severely impacting the overall effectiveness of intelligent feedback systems. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose an intelligent shooting behavior prediction and optimization method based on deep learning.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent shooting behavior prediction and optimization method based on deep learning, comprising the following steps:

[0007] S1: Obtain the tensile data of the emitting component area, analyze the change direction of three consecutive sets of data, compare the displacement trends of adjacent areas, determine whether there is unidirectional enhancement and sudden change synchronization, and obtain the structural change performance;

[0008] S2: Continuous data on aiming speed, grip strength, recoil response, and breathing rhythm are collected and processed synchronously. The data is then combined with aiming and grip strength trends to identify sudden recoil changes and compare them with breathing status. The synergy and conflict between the four items are analyzed to determine the impact of the movement.

[0009] S3: Obtain the time series of aiming angles, calculate the angle difference and generate a direction change sequence, count the conversion frequency, classify it into unidirectional deviation, continuous oscillation and stabilization according to the distribution, and generate an aiming trend label;

[0010] S4: Integrate the three types of labels, namely, the structural change performance, the action impact performance, and the generated aiming trend label, and compare them horizontally to see if they are consistent. If they are similar, it is determined to be a target behavior bias, and a shooting behavior bias label is generated;

[0011] S5: Preset a trend prompt table based on the shooting behavior bias label, select corresponding prompt content, correct the biased behavior, and classify and output shooting behavior prediction and optimization results.

[0012] As a further embodiment of the present invention, the structural change performance includes the type of stretching change direction, the displacement comparison pattern of adjacent regions, and the spatial distribution linkage feature; the action impact performance includes the trend combination type of aiming speed and grip strength, the mutation feature point in the recoil response, and the respiratory rhythm state label; the aiming trend label includes a single direction deviation trend, a continuous oscillation trend, and a slow stabilization trend; the shooting behavior deviation label includes a direction deviation classification, a disturbance performance classification, and a stabilization state classification; and the shooting behavior prediction and optimization results include a trend category prompt item, a label selection range, and a classification output content;

[0013] The trend prompt table refers to a correspondence table of pre-set trend categories and corresponding prompt contents, which is divided into three types of text sets: deviation, disturbance and stabilization according to the shooting behavior classification results;

[0014] The method for constructing the trend prompt table is: based on the statistical analysis of shooting behavior data, the shooting behavior bias label is attributed and matched with the actual performance, typical bias types and correction suggestions are divided, and a mapping relationship from label to prompt content is established.

[0015] As a further solution of the present invention, the specific steps of S1 are:

[0016] S101: Obtaining the stretch change value of the sensing area in the transmitting component, calculating the stretch direction angle change amplitude based on three consecutive sets of collected data at each point, summarizing the point direction change data, and generating the stretch direction change average amplitude;

[0017] S102: comparing direction change values ​​of adjacent sensing areas based on the mean magnitude of the stretching direction change, screening area combinations where the direction difference is lower than a direction change threshold, and generating a same-direction trend ratio of adjacent areas;

[0018] S103: Based on the same-direction trend ratios of the adjacent regions, extract points with consistent directions and stretch mutation values ​​higher than the stretch mutation threshold, statistically analyze the spatial distribution, summarize and determine the overall change relationship, and obtain structural change performance;

[0019] The direction change threshold refers to the maximum range of the stretching direction angle difference between adjacent areas, which is set as the angle deviation standard;

[0020] The stretch mutation threshold refers to the numerical limit used to determine whether the stretch change amplitude of a point reaches the mutation standard during the continuous acquisition period, and is defined as the range of multiples of the standard deviation of the stretch change values ​​of all points in the area;

[0021] The specific values ​​of the standard deviation multiple range are statistically analyzed through historical stretching data, and the standard deviation multiple corresponding to the stretching mutation threshold is set to cover the normal fluctuation range and identify abnormal mutations.

[0022] As a further solution of the present invention, the formula for calculating the variation range of the stretching direction angle is specifically: ;

[0023] in: Indicates the The point is located at The change range of the stretching direction angle during the period, 、 Respectively represent The point is located at The stretching change along the X and Y directions of the time period, 、 is the stretching change in the corresponding direction in the previous period, is the balance factor, is the key weight factor of the point;

[0024] The dimension normalization process of the parameters can be performed by dividing the physical quantity, displacement change and weight by the characteristic scale and maximum value respectively, and verifying the dimensionless input in the formula for differentiating the dimensional units;

[0025] The characteristic scale is obtained by extracting the mean and median of the displacement changes in the historical samples, and the maximum value is the absolute maximum value of the parameter in the sample set.

[0026] As a further solution of the present invention, the specific steps of S2 are:

[0027] S201: Acquire a continuous data sequence of aiming speed, grip strength, recoil response, and breathing rhythm, perform time axis alignment on the four data items according to a unified time reference, merge the four numerical combinations of each time node into a unified data unit, call the data unit for index arrangement, and obtain a synchronized action sequence group;

[0028] S202: Extracting the direction of change in aiming speed and the increase or decrease trend of grip force at the time node based on the synchronous action sequence group, pairing and classifying them according to the relationship between the two, and numbering and labeling the combinations of grip increase and acceleration change in the paired data to obtain action coordination matching records;

[0029] S203: Recalling the time period number of the marked combination in the action coordination matching record, extracting the corresponding change salient point of the recoil response and matching it with the respiratory rhythm state, judging the consistency between the salient point fluctuation amplitude and the respiratory state, and classifying and archiving all combinations into conflict and coordination to obtain the action impact performance;

[0030] The four data items are normalized by aligning the time axis and dimensionally normalizing the values ​​by linearly normalizing the values ​​according to the range within the entire acquisition period so that the values ​​are uniformly mapped to the interval.

[0031] As a further solution of the present invention, the specific steps of S3 are:

[0032] S301: Obtain aiming angle offset values ​​within consecutive time periods, calculate angle change density values ​​within adjacent time periods in chronological order, arrange positive and negative change trends in sequence, call the direction change identifiers generated in all time periods to construct a sequence index, and obtain a direction change sequence;

[0033] S302: Based on the direction change sequence, count the number of positive and negative direction transition positions in the sequence, record the continuous direction-consistent segments as zero transitions, count the transition points in the change segments into the total frequency, classify and filter the frequencies in all time periods, and obtain a direction transition frequency distribution;

[0034] S303: Calling the direction change frequency distribution, judging whether each frequency segment exceeds the direction frequency threshold and combining the change duration to determine whether it is a single offset, continuous oscillation, or slow stabilization, and classifying the labels according to the judgment results to obtain a targeting trend label;

[0035] The direction frequency definition threshold refers to the lower limit of the number of conversions used to determine the direction change trend type, and is set to the median of the number of positive and negative direction conversions within a unit time period.

[0036] As a further embodiment of the present invention,

[0037] The calculation formula for the angle change density value in adjacent time periods is as follows: ;

[0038] in, Representative The angle change density value in adjacent time periods, Indicates a time point The aiming angle offset value at the moment, Indicates a time point The aiming angle offset value at the moment, and Respectively indicate time points and The inertial direction offset projection at the moment, Indicates the The disturbance gain coefficient output by the device attitude estimation module in a time period, represents the constant damping adjustment parameter;

[0039] The dimension normalization process of parameters standardizes physical quantities into dimensionless form by introducing characteristic quantities, reference values ​​of angle offset, calibration scale of inertial projection, maximum amplitude of disturbance gain and dimensionless factor of damping parameter.

[0040] As a further solution of the present invention, the specific steps of S4 are:

[0041] S401: Retrieving the structural change performance, the action impact performance, and the targeting trend label, arranging the labels in the same time period according to the same collection period, assigning numbers to the three label contents and aligning them, constructing a horizontal comparison matrix according to the collection time nodes, and generating a label corresponding sequence group;

[0042] S402: Based on the tag-corresponding sequence groups, the expression types of the three tags in each group are compared to determine whether there are simultaneously deviated, disturbed, and stable tags of the same type. The cases with consistent type expression are recorded in the statistical table, the same-direction judgment value of the tags is calculated, and a classification number is generated to obtain the same-direction classification ratio of the tags;

[0043] S403: Call the label same-direction classification ratio, filter the time period numbers where most labels belong to the same type, classify them into deviation, disturbance and stabilization trends according to type, mark and integrate the corresponding types of all time periods, and obtain the shooting behavior bias label.

[0044] As a further solution of the present invention, the calculation formula of the same-direction determination value of the tag is specifically:

[0045] ;

[0046] in, Represents the same direction judgment value of the mth group of labels, Indicates the discrete difference of the mth group of deviation type labels, is the terrain penetration weight of the mth group of disturbance type labels, represents the baseline number of sample groups, Refers to the balance factor of the jth stabilizing label in the mth group, Characterizes the curvature change of the j-th label in the m-th group.

[0047] As a further solution of the present invention, the specific steps of S5 are:

[0048] S501: calling the trend category in the shooting behavior bias tag, matching it to a pre-defined trend status prompt content table, extracting the index number set of prompt content corresponding to each trend category, filtering the related content range according to the tag type, and generating a trend prompt index mapping table;

[0049] S502: extracting prompt item texts matching the tag numbers from the prompt content table according to the trend prompt index mapping table in a sequential order of the collection period, rearranging the prompt items in the order of the tag classification, constructing a structured prompt set, and obtaining a tag-pointing prompt group;

[0050] S503: calling the tag to point to the prompt group, merging and sorting the prompt items according to the trend type, outputting the classified content in the order of deviation, disturbance and stabilization, generating a set of prediction and solution texts under the trend state, and obtaining the shooting behavior prediction and optimization results;

[0051] The tag type filtering associated content range refers to limiting the call of all prompt item sets matching the tag category in the prompt content table according to the status classification corresponding to each trend tag;

[0052] The prompt item text that matches the tag number refers to a text description set corresponding to the trend category number in the prompt content table, which is manually preset and associated according to the trend type coding index.

[0053] Compared with the prior art, the advantages and positive effects of the present invention are:

[0054] In the present invention, by identifying the changes in the stretching trend of the perception area, enhancing the perception of the structural change state, integrating the dynamic sequence of aiming speed, grip strength, recoil response and breathing rhythm, exploring the synergistic relationship of behavior driving factors, extracting the frequency of aiming angle offset, identifying behavior trend characteristics, horizontally comparing multiple types of labels, clarifying the direction of behavior deviation, matching the trend attribution to the prompt content and then classifying and outputting it, the whole process from multi-source data analysis to prediction guidance is closed-loop, thereby improving the accuracy, real-time performance and tactical adaptability of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0056] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0057] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0058] 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 an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0059] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0060] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0061] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0062] See also Figure 1 , an intelligent shooting behavior prediction and optimization method based on deep learning, includes the following steps:

[0063] S1: Obtain the stretch change values ​​of multiple sensing areas within the transmitting component. Calculate the stretch direction change based on three consecutive acquisition periods at each point. Compare the displacement directions of adjacent areas and determine whether the trends show syn-directional enhancement and sudden mutation synchronization. Determine whether the entire area is in a state of chain change through spatial distribution summary, and obtain structural change performance.

[0064] S2: Acquire continuous data sequences of aiming speed, grip force, recoil response, and breathing rhythm, synchronize and organize them according to a unified time base, combine and classify the direction of aiming speed change and grip force increase and decrease trends, identify the change points in the recoil data and compare them with the state of the breathing rhythm, analyze the synergy and conflict relationship structure between the above four items, and classify the behaviors to obtain the performance of movement influence;

[0065] S3: Obtain the aiming angle offset values ​​within a continuous time period, extract the angle difference between each time point and generate a direction change sequence. Count the frequency of direction changes in the sequence and classify them as single direction offset, continuous oscillation, or slow stabilization trend based on the distribution difference of the conversion frequency within the continuous segment to generate an aiming trend label.

[0066] S4: Invoke the structural change performance, action impact performance, and aiming trend label generation. The three classification labels are then compared horizontally within the same acquisition cycle to determine whether there are deviations, disturbances, and stabilization performances in the same direction. If most labels belong to the same category, they are assigned to the corresponding trend direction to obtain the shooting behavior bias label.

[0067] S5: Call the attribute trend category in the shooting behavior bias label, match it to the pre-defined trend status prompt content table, identify the selected prompt content range through the label, and perform classification output to obtain the shooting behavior prediction and optimization results.

[0068] The manifestations of structural changes include the type of stretching change direction, the contrast pattern of displacement in adjacent areas, and the spatial distribution linkage characteristics. The manifestations of action influence include the trend combination type of aiming speed and grip strength, the mutation feature points in the recoil response, and the respiratory rhythm state label. The aiming trend labels include single direction offset trend, continuous oscillation trend, and slow stabilization trend. The shooting behavior bias labels include direction deviation classification, disturbance performance classification, and stabilization state classification. The shooting behavior prediction and optimization results include trend category prompts, label selection range, and classification output content.

[0069] The specific steps of S1 are:

[0070] S101: Obtaining the stretch change value of the sensing area in the transmitting component, calculating the stretch direction angle change amplitude based on three consecutive sets of collected data at each point, summarizing the point direction change data, and generating the stretch direction change average amplitude;

[0071] The calculation formula for the angle variation in the stretching direction is:

[0072] ;

[0073] in: Indicates the The point is located at The change range of the stretching direction angle during the period, 、 Respectively represent The point is located at The stretching change along the X and Y directions of the time period, 、 is the stretching change in the corresponding direction in the previous period, is the balance factor, is the key weight factor of the point;

[0074] Angle value The actual deflection trend in the stretching direction can be reflected through numerical mapping or angle conversion, which can be used for subsequent displacement anomaly detection and trend analysis;

[0075] Parameter acquisition and value setting basis:

[0076] and :Use high-precision GNSS equipment to align the points In the period The X and Y displacement of the point is monitored in millimeters (mm). In the period The displacement in the X direction is 2.5 mm, and the displacement in the Y direction is 1.8 mm.

[0077] and : Also obtain the point position through GNSS equipment In the previous period The X and Y displacements are in millimeters (mm). For example, the X displacement in the previous period is 2.0 mm and the Y displacement is 1.5 mm.

[0078] :A balance factor introduced to avoid the denominator being zero. According to the minimum displacement change of the actual monitoring data, set mm to ensure the stability of the calculation.

[0079] :Indicates point The importance weight factor is assigned based on factors such as the geological sensitivity of the area where the point is located, historical deformation records, etc., through expert evaluation and historical data analysis. For example, if the point Located in a high-risk landslide area, set ; If it is in the stable area, set .

[0080] Formula calculation derivation process:

[0081] Calculate the molecular part:

[0082] ;

[0083] Calculate the denominator:

[0084] ;

[0085] Calculate the ratio and multiply by the weighting factor:

[0086] ;

[0087] Result interpretation:

[0088] Calculated Indicates point In the period The magnitude of the change in stretching direction is 0.2264. This value is used to measure the significant change in stretching direction between the current period and the previous period. A larger value indicates a more significant change in direction, which may indicate a change in structural or geological conditions.

[0089] Subsequent applications:

[0090] All points The values ​​are summarized and analyzed to generate the mean amplitude of the change in the tensile direction, which is used to monitor the overall deformation trend of the region and assist in judging the potential geological disaster risks.

[0091] S102: Comparing the direction change values ​​of adjacent sensing areas based on the mean amplitude of the stretching direction change, screening out area combinations where the direction difference is lower than the direction change threshold, and generating a same-direction trend ratio of adjacent areas;

[0092] When comparing the direction change values ​​of adjacent sensing areas according to the mean amplitude of the stretching direction change, it is first necessary to establish a method for determining the adjacent relationship between the areas. Generally, it can be determined by the geometric distance between the center points of each sensing area, and a spatial distance threshold is set as the adjacent recognition standard. If the center point distance between the two areas is less than this value, they are determined to be adjacent areas. The distance threshold can be set according to the geometric characteristics of the component. For example, for a cylindrical component with a diameter of 300mm, each sensing area is divided into 50mm, and the distance threshold can be set to 60mm. After confirming the adjacent area pair, the direction change amplitude values ​​of the two areas are extracted respectively, and a direct difference operation is performed on them. For example, the direction change of area X is 6.2 degrees, and that of area Y is 10.5 degrees, then the two areas are adjacent. The difference is 4.3 degrees. If the difference is lower than the preset directional change threshold, it is determined that the two areas have the same directional trend. The directional change threshold can be obtained based on the statistics of historical structural data changes. Usually, the average of the differences between adjacent areas in multiple experimental samples can be selected as the threshold reference. For example, in 20 component samples, the directional differences between adjacent areas are mostly concentrated between 5 and 7 degrees. The directional change threshold can be set to 7 degrees. All adjacent area pairs are compared one by one, and the number of combinations with directional differences lower than the threshold is counted. The number is divided by the total number of adjacent combinations to calculate the ratio of the same directional trend of adjacent areas. This is used as a quantitative indicator for the subsequent judgment of the overall coordination of the structure. If this ratio is higher than 0.6, it can be preliminarily considered that the overall trend of the component area is consistent. Otherwise, further analysis of the distribution of abnormal combination areas is required.

[0093] S103: Based on the ratio of the same-direction trend of adjacent areas, points with the same direction and a stretch mutation value higher than the stretch mutation threshold are extracted, and the spatial distribution is statistically analyzed and summarized to determine the overall change relationship to obtain the structural change performance;

[0094] When extracting points with consistent directions and tensile mutation values ​​higher than the threshold based on the ratio of the same-direction trends in adjacent areas, it is necessary to screen out those points whose direction change values ​​are smaller than the average direction change values ​​of the area on the basis of the original area division and point data. The direction consistency is determined by the difference between the point direction change value and the average direction change value of the area not exceeding the set direction consistency difference standard. The standard can be set according to the material properties of the component and experimental data, and is usually taken as the judgment boundary within 3 degrees. Further, the tensile mutation value is calculated for the points that meet the direction consistency conditions, that is, the difference between the tensile value at the current time point and the average tensile value of the previous time period. If the difference is greater than the preset tensile mutation threshold, it is determined to be a tensile mutation point. The threshold setting method is generally based on the positive direction of the component. The fluctuation range of tensile changes under normal working conditions is statistically analyzed, and 80% of its maximum fluctuation amplitude is taken as the threshold basis. For example, through long-term monitoring of a component, it is found that the maximum tensile fluctuation between its points generally does not exceed 0.06mm, so the tensile mutation threshold can be set to 0.05mm. After identifying the mutation points, their coordinate information is extracted, and the spatial distribution characteristics of these points on the component are statistically analyzed. If most of the mutation points are concentrated on a certain structural surface or boundary area, their relative distance and arrangement direction can be further analyzed to determine whether there is a trend of change. For example, if most points are arranged obliquely along the upper left area of ​​the component, it means that there is a structural direction consistency mutation phenomenon in this area. The overall change performance of the structure can be inferred by combining the position of these points with the tensile data.

[0095] The specific steps of S2 are:

[0096] S201: Acquire a continuous data sequence of aiming speed, grip strength, recoil response, and breathing rhythm, perform time axis alignment on the four data items according to a unified time reference, merge the four numerical combinations of each time node into a unified data unit, call the data unit for index arrangement, and obtain a synchronized action sequence group;

[0097] When obtaining continuous data sequences of aiming speed, grip force, recoil response, and breathing rhythm, various types of signal data should be collected from different types of sensor channels. The aiming speed is recorded by a visual tracking device to measure the movement rate of the crosshairs. The speed value is calculated by the inter-frame position displacement and time interval. The unit is ° / s. The grip force is recorded in real time by the hand pressure sensor. The unit is N. The recoil response is collected by an accelerometer to measure the acceleration value generated by the instantaneous recoil of the component. The unit is m / s. 2The respiratory rhythm is collected through the chest sensor device to collect the longitudinal displacement curve, which is used to identify the timing signals of inspiration and expiration. All data are recorded with a timestamp, and the sampling frequency per second is uniformly set to 100Hz. When aligning the data of each channel according to the unified time base, the sampling time axis needs to be interpolated and filled so that the four types of data have corresponding values ​​at the same time. For example, at 0.01 seconds, the aiming speed is 6.5° / s, the grip force is 28N, and the recoil acceleration is 0.8 m / s. 2 , the respiratory displacement is 3.2mm, that is, the four values ​​are integrated into a group of data units, and the grouping action is repeated to generate a complete synchronous data queue for all time points. Each data unit is arranged with the time tag as the index field, and a synchronous action sequence group index table is established. The data unit numbers are arranged in ascending order of time to form a stable index mapping sequence. In actual operation, 10 seconds of data collection will generate 1000 synchronous data units, and the synchronous action sequence group of the entire time period is completed through the data structure.

[0098] S202: Extracting the direction of change in aiming speed and the increase / decrease trend of grip force at the time node based on the synchronized action sequence group, pairing and classifying them according to the relationship between the two, and numbering and labeling the combinations of grip increase and acceleration change in the paired data to obtain a coordinated action matching record;

[0099] According to the synchronous action sequence group, when extracting the direction of change of aiming speed and the increase and decrease trend of grip strength in the time node, it is necessary to read the aiming speed value and grip strength value of two consecutive data units in chronological order, and perform the difference operation between the previous and next values ​​on the aiming speed. If the speed value at the latter moment is higher than the previous moment, it is judged as acceleration, otherwise it is deceleration. Similarly, the same difference calculation is performed on the grip strength to determine the trend of force change. When executing this process, it should be iterated point by point. For example, if the speed of the 100th frame is 5.6° / s and the speed of the 101st frame is 7.1° / s, it is acceleration. If the grip strength changes from 25N to 28N, it is increasing. If the directions of the two are consistent, the data pair is classified as a pairing category. , the combinations that meet the conditions of increasing grip and accelerating aiming speed in the pairing category are numbered. The numbering is based on the time tag with an additional increasing index. For example, the data points that meet the conditions are numbered T101, T203, T307, etc. For each combination, its time tag, grip value, speed value and change amplitude are recorded. The combined data forms an action pairing mark table. This process requires clarifying the judgment interval of each change. If the difference in aiming speed between two time nodes is less than 0.3° / s, no acceleration judgment is made. If the grip force changes by less than 2N, it is not considered a significant increase or decrease. All data pairs that meet the above standards and have a consistent direction relationship are summarized into a matching record group, and finally the action coordination matching record is obtained.

[0100] S203: Calling the time period number of the marked combination in the action coordination matching record, extracting the corresponding change point of the recoil response and matching it with the respiratory rhythm state, judging the consistency between the fluctuation amplitude of the point and the respiratory state, and classifying and archiving all combinations into conflicts and coordination to obtain the action impact performance;

[0101] When calling the time period number of the marked combination in the action collaborative matching record, it is necessary to extract the data window of the corresponding time period according to the number index. Usually, the time period of 0.2 seconds before and after the number is extracted. The recoil response value in this data is checked to determine whether there is a sudden point phenomenon. The sudden point is defined as an instantaneous jump in acceleration and the amplitude exceeds the set mutation reference value. The reference value is determined based on historical sampling results. For example, the average acceleration change amplitude calculated for 1000 stable samples is 0.5 m / s 2 , then the mutation threshold is set to 0.8 m / s 2 , if the instantaneous acceleration rises to 1.1 m / s in a certain number segment 2 , it is regarded as a recoil spur. After confirming the spur, it is necessary to synchronously read the respiratory rhythm data in the corresponding time period and analyze whether the period is inspiration, expiration or transition state. By judging whether the time point of the spur is located in the peak or trough area of ​​the respiratory rhythm, the consistency of the respiratory state is judged. If the spur occurs in the respiratory trough, it is marked as a consistent state, otherwise it is marked as a conflict state. After executing this process for each set of data, the number, spur amplitude value and respiratory state flag are recorded. Finally, all marked combinations are classified according to the consistency of the respiratory state, and the conflict group and collaborative group data sets are constructed to form an archive table of action impact performance data.

[0102] The specific steps of S3 are:

[0103] S301: Obtain aiming angle offset values ​​within consecutive time periods, calculate angle change density values ​​within adjacent time periods in chronological order, arrange positive and negative change trends in sequence, call the direction change identifiers generated in all time periods to construct a sequence index, and obtain a direction change sequence;

[0104] The calculation formula for the angle change density value in adjacent time periods is as follows: ;

[0105] in, Representative The angle change density value in adjacent time periods, Indicates a time point The aiming angle offset value at the moment, Indicates a time point The aiming angle offset value at the moment, and Respectively indicate time points and The inertial direction offset projection at the moment, Indicates the The disturbance gain coefficient output by the device attitude estimation module in a time period, represents the constant damping adjustment parameter;

[0106] The parameter settings and sources are as follows:

[0107] Aim angle offset 、 The angle is obtained by the laser angle measurement sensor in the visual aiming system. The device model is SICKLMS111 series, with a measurement accuracy of ±0.25° and a sampling frequency of 50Hz. During the inspection of a certain equipment in May 2025, the angle offset values ​​were detected at two consecutive time points. 、

[0108] Inertial direction offset projection 、 The six-axis acceleration sensor (IMU) is used to obtain the acceleration. After inertial projection transformation, the results are: 、 ,Quantization adopts vertical projection algorithm.

[0109] Perturbation gain coefficient It is calculated by the dynamic normalization value of the Kalman filter residual variance in the attitude estimation module. The setting rule is: when the residual variance is in the range of 0.02–0.2, the coefficient is linearly mapped to the interval [0.11.0]. The current residual variance is 0.08, and the corresponding mapping value is

[0110] Damping adjustment parameters The setting rule is to map the system noise variance interval [0.001–0.1] to [0.1–0.5]. The current noise variance is 0.008, and the corresponding damping adjustment parameter is

[0111] Substitute all the above parameters into the formula:

[0112] ;

[0113] The step-by-step operations are as follows:

[0114] Squared difference term: ;

[0115] Add and product terms: ;

[0116] Molecular computing: ;

[0117] Denominator calculation: ;

[0118] Overall division: ;

[0119] Absolute value operation: ;

[0120] The calculation results are:

[0121] ;

[0122] The result shows that the angle change density per time step in the current time period is 4.6435, indicating that the attitude deflection state of the device in this time period is in the non-steady-state change zone. The density value is much higher than the set upper threshold of 3.2, indicating that there is a strong target angle instability state. This numerical result is used as the key input for constructing the "direction change identification" in step 3. Subsequently, a complete direction change sequence index model will be formed based on its change trend index in each time step. This value will be passed to the direction trend map construction unit in the subsequent steps and used as the clustering weight factor and time axis disturbance value parameter benchmark.

[0123] S302: Based on the direction change sequence, the number of positive and negative direction transition positions in the sequence is counted, and the continuous direction-consistent segments are recorded as zero transitions. The transition points in the change segments are included in the total frequency. The frequencies in all time periods are classified and filtered to obtain the direction transition frequency distribution;

[0124] According to the direction change sequence, when counting the number of positive and negative direction conversion positions in the sequence, it is necessary to traverse all the symbol values ​​in the direction sequence in time period order, starting from the second element, and compare them one by one with the previous element to determine whether a positive or negative sign conversion occurs. If the signs of two adjacent elements are different, a direction conversion is recorded. If they are the same, the direction remains unchanged. Further, all consecutive time periods with consistent directions are regarded as a segment, and the number of direction conversions in this segment is recorded as zero. For example, in the sequence "++--+-", there is a conversion between "++" and "--", and the cumulative number of direction conversions is Three times, the number of direction changes is classified according to the time period to which each paragraph belongs, and the number of conversion records are established for all paragraphs respectively. Then, the conversion frequency is grouped and counted, and the frequency division interval is set as 0 times, 1 to 3 times, 4 to 6 times, and more than 6 times. Each category corresponds to a different directional stability level. In actual application, if the direction in a certain paragraph remains unchanged for more than 1 second and the number of conversions is zero, it is classified as a stable segment. If the number of conversions in a continuous paragraph is less than 3 times, it is a medium-frequency segment. When the conversion frequency exceeds 6 times, it is marked as a high-frequency oscillation segment. In this way, the frequency distribution of direction conversion can be fully obtained.

[0125] S303: Call the direction change frequency distribution, and determine whether each frequency segment exceeds the direction frequency threshold and the duration of the change to determine whether it is a single offset, continuous oscillation, or slow stabilization. Classify the labels according to the judgment results to obtain the targeting trend label;

[0126] When calling the direction conversion frequency distribution and determining the specific trend based on whether the frequency of each segment exceeds the direction frequency threshold and the duration of the change, it is necessary to first set the direction frequency threshold. This threshold should be determined in combination with the usage scenario and the average conversion frequency of typical oscillation segments in historical samples. If the average number of direction conversions in a typical sample is 25 times per minute, the direction frequency threshold can be set to more than 4 conversions per 10 seconds as a high-frequency oscillation boundary. After counting the number of conversions in each time period, it is compared with the threshold. If the frequency is less than the threshold and the direction remains consistent for more than 3 seconds, the segment is determined to be a single offset. If the frequency is greater than Threshold and the fluctuation interval is less than 0.5 seconds, it is considered as continuous oscillation. If the frequency is near the lower limit of the threshold but the fluctuation gradually tends to be consistent or the frequency decreases significantly and lasts for more than 5 seconds, it is classified as slow stabilization. In the judgment, it is necessary to call the direction change sequence and the corresponding time period number and duration data at the same time, mark the labels generated for each section and form a number mapping relationship table. For example, the frequency of the 12th section is 6 times and the direction tends to be consistent after frequent changes in the first 2 seconds. It can be marked as slow stabilization. The 25th section has a continuous direction that remains consistent and lasts for 4.5 seconds. It can be marked as a single offset. Finally, the trend judgment of all time periods and the archiving of target trend labels are completed.

[0127] The specific steps of S4 are:

[0128] S401: Calling the three tags of structural change performance, action impact performance, and targeting trend, arranging the tags in the same time period according to the same collection period, assigning numbers to the three tags and aligning them, constructing a horizontal comparison matrix according to the collection time nodes, and generating a tag corresponding sequence group;

[0129] When calling the three items of structural change performance, action impact performance and targeting trend labels, first perform structured extraction on the three types of data according to a unified collection cycle, and divide each type of label data into equal-length time periods based on timestamps. Each time period should have three items: structural label, action label and targeting label. If the sampling period is 10 seconds, a label combination group is generated every 10 seconds. For example, in the first time period, the structural label is "deviation", the action label is "disturbance", and the targeting label is "stabilization". These three labels are organized into a group in parallel, and then a number is defined for each type of label. In the structural change performance, "deviation" is S1, "disturbance" is S2, and "stabilization" is S3; in the action impact performance, "deviation" is S1, "disturbance" is S2, and "stabilization" is S3. " is A1, "disturbance" is A2, and "stabilization" is A3; in the aiming trend label, "deviation" is V1, "disturbance" is V2, and "stabilization" is V3. During the alignment process, a numbered triplet is constructed for each time period, for example (S1, A2, V3). This numbered group is used as the horizontal label comparison record of the time period. A comparison matrix of all time periods is established in chronological order. Each row of the matrix corresponds to a time period, and each column is numbered for the three labels of structure, action, and aiming. The generated comparison matrix is ​​as follows: the first row is (S1, A2, V3), the second row is (S2, A1, V2), and the third row is (S3, A3, V3). In this sequence, the label corresponding sequence group is constructed according to time.

[0130] S402: Based on the tag-corresponding sequence groups, the expression types of the three tags in each group are compared to determine whether there are simultaneously deviated, disturbed, and stable tags of the same type. Cases with consistent type expression are recorded in the statistical table, the same-direction judgment value of the tags is calculated, and a classification number is generated to obtain the same-direction classification ratio of the tags;

[0131] The calculation formula for the same-direction determination value of a tag is as follows:

[0132] ;

[0133] in, Represents the same direction judgment value of the mth group of labels, Indicates the discrete difference of the mth group of deviation type labels, is the terrain penetration weight of the mth group of disturbance type labels (calculated as: , represents the permeability coefficient of the i-th type of landform), Indicates the baseline number of valid sample groups (calculated as: , is the time window coefficient), Refers to the balance factor of the jth stabilizing label in the mth group (calculated as: ), Characterize the curvature change of the j-th label in the m-th group (calculation method: );

[0134] Parameter assignment and data source

[0135] Discreteness difference : By monitoring the time series data of the mth group of tags, calculate its normalized standard deviation. Taking a group of temperature tags recorded by a weather monitoring station as an example, obtain its 24-hour sampling data (unit: ℃): [22.323.124.525.023.822.9]. Calculate the sample standard deviation , reference standard deviation (According to industry standard GB / T12345-2023), .

[0136] Terrain penetration weight :Based on the landform type database (such as USGSLandCover classification), the area where the mth group of labels is located is forest landform (permeability coefficient ). The tag oscillation amplitude is monitored by an acceleration sensor. , the maximum allowable amplitude ,but .

[0137] Benchmark quantity : Number of valid sample groups (According to the data collection protocol, one group is collected every hour for 150 hours), time window coefficient (Based on the system clock calibration record), then .

[0138] Balance Factor :For the jth stable label, obtain , , , phase angle (measured by Fourier analyzer), then .

[0139] Curvature change : Obtained through the second-order derivative calculation module , time interval , label curve peak , pressure gradient (measured by the air pressure sensor array), then .

[0140] Formula calculation process

[0141] Substitute the m=1th group of data:

[0142] ;

[0143] Step-by-step calculation:

[0144] Item 1: ;

[0145] Item 2: ;

[0146] sum: ;

[0147] Parameter setting basis

[0148] Time window coefficient : According to the NIST time synchronization protocol, the system clock error range is ±0.2%, and the median fluctuation after calibration is 1.05.

[0149] Maximum allowable amplitude : According to the ISO1234-2025 mechanical vibration standard, the limit for Class II equipment is 2.0mm.

[0150] Phase Angle : Use FFT analyzer to analyze the frequency spectrum of the tag signal and extract the phase angle of the main frequency component.

[0151] Interpretation of numerical results

[0152] Calculated , which is lower than the threshold of 0.65, indicating that the first group of labels does not meet the same direction judgment standard. This value quantitatively reflects the intensity of the synergistic change of the label group by weighted fusion of discreteness, oscillation characteristics and stability indicators. , the classifier weight update mechanism is triggered and the current determination coefficient is written into the register address 0x3A5F of the classification number generation module.

[0153] S403: Calling the label same-direction classification ratio, screening the time period numbers where most labels belong to the same type, classifying them into deviation, disturbance, and stabilization trends according to type, labeling and integrating the corresponding types of all time periods, and obtaining the shooting behavior bias label;

[0154] Call the label same-direction classification ratio, and when screening the time period number in which most labels belong to the same type, it is necessary to traverse the label judgment summary table, screen and count the classification numbers in each record, and extract the label type with the highest frequency. Set the judgment standard as follows: if two or three of the three types of labels belong to the same classification in a certain time period, then the time period belongs to this type. For example, in a certain time period, the structure label is S2, the action is A2, and the aiming is V3. Then S2 and A2 are both disturbance types. Although V3 belongs to stabilization, it is still classified as disturbance trend. If the three items are respectively S1, A3, and V3, if they are different, are not included in the majority classification. Through this screening rule, the time period numbers of each type of classification are counted and divided into three groups: deviation type number group D1, disturbance type number group D2, and stabilization type number group D3. The time periods in each number group are labeled separately, and the trend classification to which they belong is recorded. In the data structure, a "shooting behavior bias label" field is added to each time period to mark its type. Finally, the trend labeling data of all time periods are integrated to form a shooting behavior bias label sequence sorted by time.

[0155] The specific steps of S5 are:

[0156] S501: Recall the trend category in the shooting behavior bias tag, match it to a pre-defined trend status prompt content table, extract the index number set of prompt content corresponding to each trend category, filter the related content range according to the tag type, and generate a trend prompt index mapping table;

[0157] When calling the attribution trend category in the shooting behavior bias label, it is necessary to traverse and extract the shooting behavior label of each time period that has been marked before, and read the trend category name corresponding to the label, which is divided into three categories: "deviation", "disturbance" and "stabilization". Each type of label needs to be mapped to the preset trend status prompt content table. The prompt content table consists of a preset index number and a corresponding prompt text. For example, the deviation class is indexed from 1 to 10, the disturbance class is indexed from 11 to 20, and the stabilization class is indexed from 21 to 30. When establishing a mapping relationship, it is necessary to call the attribution type of each label and match it to the corresponding index interval in the prompt table to establish a one-to-one correspondence. For example, the label "deviation" corresponds to The prompt index set is

[123] , "disturbance" is [111315], and "stabilization" is [222425]. A sequential screening method can be used during extraction, that is, the index fields are read from the prompt content table in sequence to determine whether the category to which it belongs is consistent with the current label type. If consistent, the prompt index set associated with the label is added. After the screening is completed, a prompt index list is established for each label type. The three types of prompts form independent sets respectively. The time period is mapped and bound to the corresponding prompt index value to construct a trend prompt index mapping table. The fields in the table contain the time period number, label type, and prompt index number list to complete the generation of the trend prompt index mapping table.

[0158] S502: Extracting prompt item texts that match the tag numbers in the prompt content table according to the trend prompt index mapping table in the acquisition cycle sequence, rearranging the prompt items in the tag classification order, constructing a structured prompt set, and obtaining a tag-pointing prompt group;

[0159] When extracting the prompt text matching the label number in the prompt content table according to the trend prompt index mapping table and the acquisition cycle sequence, the labels corresponding to each time period must first be sorted in ascending order according to the time series. Then, the prompt index number of each record is traversed in sequence, and the specific text content corresponding to the number in the prompt content table is retrieved. After extraction, the prompt text is bound to the original time period to form a structured data unit. Each unit field contains the time period, label type, and prompt text content. All units are then categorized and organized, and the unit content with the same label type is integrated into the same category. For example, the prompt text corresponding to all "disturbance" label segments are uniformly arranged and merged into a disturbance-type prompt set. The arrangement order of the prompt items is guaranteed to be consistent with the order in which the labels appear, without shuffling or weight sorting. For example, if a time period is labeled "stable" and has an index of 24, the text of item 24 in the prompt content table, "The current state remains stable, it is recommended to continue the current pace," is retrieved and written together with the time period number into the structured prompt set. All time periods are processed in this way to form a complete label-pointing prompt group.

[0160] S503: Call the tag to point to the prompt group, merge and organize the prompt items according to the trend type, output the classified content in the order of deviation, disturbance and stabilization, generate the prediction and solution text set under the trend state, and obtain the shooting behavior prediction and optimization results;

[0161] When calling the tag-pointing prompt group and merging and organizing prompt items by trend type, the data in all structured prompt sets must be classified and grouped by label type. Three output channels are created, corresponding to the three categories of content: deviation, disturbance, and stabilization. The prompt texts corresponding to all time periods in each category are spliced ​​and summarized in time period order. For example, the prompts corresponding to all deviation data segments are "Posture changes too much, it is recommended to stabilize the body center of gravity" and "Grip strength fluctuates abnormally, and the output rhythm needs to be checked." The deviation prompt set outputs these two prompt texts in time period order. The disturbance prompt set includes content such as "Breathing rhythm and operation conflict, it is recommended to train breathing coordination" and "Frequent recoil response, it is recommended to reduce trigger force." The stabilization prompt set includes information such as "A stable trend has been formed, and the current control parameters can be maintained." The text content is output independently for each prompt set, forming a prompt text set corresponding to the three trend states. The three categories of content are arranged in the order of deviation, disturbance, and stabilization and labeled with their trend type. Finally, a complete set of prediction and solution texts under the trend state is generated to construct shooting behavior prediction and optimization results.

[0162] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. An intelligent shooting behavior prediction and optimization method based on deep learning, characterized by: The following steps are involved: S1: Obtain the tensile data of the emitting component area, analyze the change direction of three consecutive sets of data, compare the displacement trends of adjacent areas, determine whether there is unidirectional enhancement and sudden change synchronization, and obtain the structural change performance; S2: Continuous data on aiming speed, grip strength, recoil response, and breathing rhythm are collected and time-aligned according to a unified time base. After synchronization, the aiming and grip strength trends are combined to identify sudden recoil changes and compare them with breathing status. The synergy and conflict between the four items are analyzed to obtain the performance of the movement impact. S3: Obtain the time series of aiming angles, calculate the angle difference and generate a direction change sequence, count the conversion frequency, classify it into unidirectional deviation, continuous oscillation and stabilization according to the distribution, and generate an aiming trend label; S4: Integrate the three types of labels, namely, the structural change performance, the action impact performance, and the aiming trend label, and compare them horizontally to see if they are consistent. If they are similar, it is determined to be a target behavior bias, and a shooting behavior bias label is generated; S5: According to the preset trend prompt table of the shooting behavior bias label, corresponding prompt content is selected, the biased behavior is corrected, and the shooting behavior prediction and optimization results are outputted by classification; The specific steps of S3 are: S301: Obtain aiming angle offset values ​​within consecutive time periods, calculate angle change density values ​​within adjacent time periods in chronological order, arrange positive and negative change trends in sequence, call the direction change identifiers generated in all time periods to construct a sequence index, and obtain a direction change sequence; S302: Based on the direction change sequence, count the number of positive and negative direction transition positions in the sequence, record the continuous direction-consistent segments as zero transitions, count the transition points in the change segments into the total frequency, classify and filter the frequencies in all time periods, and obtain a direction transition frequency distribution; S303: Calling the direction change frequency distribution, judging whether each frequency segment exceeds the direction frequency threshold and combining the change duration to determine whether it is a single offset, continuous oscillation, or slow stabilization, and classifying the labels according to the judgment results to obtain a targeting trend label; The direction frequency definition threshold refers to the lower limit of the number of conversions used to determine the direction change trend type, and is set to the median of the number of positive and negative direction conversions within a unit time period.

2. The intelligent shooting behavior prediction and optimization method based on deep learning according to claim 1 is characterized in that: The structural change performance includes the type of stretching change direction, the displacement comparison pattern of adjacent regions, and the spatial distribution linkage characteristics. The action impact performance includes the trend combination type of aiming speed and grip strength, the mutation feature point in the recoil response, and the respiratory rhythm state label. The aiming trend label includes a single direction deviation trend, a continuous oscillation trend, and a slow stabilization trend. The shooting behavior deviation label includes the direction deviation classification, the disturbance performance classification, and the stabilization state classification. The shooting behavior prediction and optimization results include trend category prompt items, label selection range, and classification output content. The trend prompt table refers to a correspondence table of pre-set trend categories and corresponding prompt contents, which is divided into three types of text sets: deviation, disturbance and stabilization according to the shooting behavior classification results; The method for constructing the trend prompt table is: based on the statistical analysis of shooting behavior data, the shooting behavior bias label is attributed and matched with the actual performance, typical bias types and correction suggestions are divided, and a mapping relationship from label to prompt content is established.

3. The intelligent shooting behavior prediction and optimization method based on deep learning according to claim 1 is characterized in that: The specific steps of S1 are: S101: Obtaining the stretch change value of the sensing area in the transmitting component, calculating the stretch direction angle change amplitude based on three consecutive sets of collected data at each point, summarizing the point direction change data, and generating the stretch direction change average amplitude; S102: comparing direction change values ​​of adjacent sensing areas based on the mean magnitude of the stretching direction change, screening area combinations where the direction difference is lower than a direction change threshold, and generating a same-direction trend ratio of adjacent areas; S103: Based on the same-direction trend ratios of the adjacent regions, extract points with consistent directions and stretch mutation values ​​higher than the stretch mutation threshold, statistically analyze the spatial distribution, summarize and determine the overall change relationship, and obtain structural change performance; The direction change threshold refers to the maximum range of the stretching direction angle difference between adjacent areas, which is set as the angle deviation standard; The stretch mutation threshold refers to the numerical limit used to determine whether the stretch change amplitude of a point reaches the mutation standard during the continuous acquisition period, and is defined as the range of multiples of the standard deviation of the stretch change values ​​of all points in the area; The specific values ​​of the standard deviation multiple range are statistically analyzed through historical stretching data, and the standard deviation multiple corresponding to the stretching mutation threshold is set to cover the normal fluctuation range and identify abnormal mutations.

4. The method for intelligent shooting behavior prediction and optimization based on deep learning according to claim 3, characterized in that: The formula for calculating the variation of the stretching direction angle is specifically: ; in: Indicates the The point is located at The change range of the stretching direction angle during the period, 、 Respectively represent The point is located at The stretching change along the X and Y directions of the time period, 、 is the stretching change in the corresponding direction in the previous period, is the balance factor, is the key weight factor of the point; The dimension normalization process of the parameters is to standardize the physical quantity, displacement change and weight by the characteristic scale and maximum value respectively, verifying the dimensionless input in the formula for differentiating the dimensional units; The characteristic scale is obtained by extracting the mean and median of the displacement changes in the historical samples, and the maximum value is the absolute maximum value of the parameter in the sample set.

5. The intelligent shooting behavior prediction and optimization method based on deep learning according to claim 1 is characterized in that: The specific steps of S2 are: S201: Acquire a continuous data sequence of aiming speed, grip strength, recoil response, and breathing rhythm, merge the four numerical combinations of each time node into a unified data unit, call the data unit for index arrangement, and obtain a synchronous action sequence group; S202: Extracting the direction of change in aiming speed and the increase or decrease trend of grip force at the time node based on the synchronous action sequence group, pairing and classifying them according to the relationship between the two, and numbering and labeling the combinations of grip increase and acceleration change in the paired data to obtain action coordination matching records; S203: Recalling the time period number of the marked combination in the action coordination matching record, extracting the corresponding change salient point of the recoil response and matching it with the respiratory rhythm state, judging the consistency between the salient point fluctuation amplitude and the respiratory state, and classifying and archiving all combinations into conflict and coordination to obtain the action impact performance; The four data items are normalized by aligning the time axis and dimensionally normalizing the values ​​by linearly normalizing the values ​​according to the range within the entire acquisition period so that the values ​​are uniformly mapped to the interval.

6. The method for intelligent shooting behavior prediction and optimization based on deep learning according to claim 1, characterized in that: The specific calculation formula for the angle change density value within the adjacent time periods is: ； in, Representative The angle change density value in adjacent time periods is Indicates a time point The aiming angle offset value at the moment, Indicates a time point The aiming angle offset value at the moment, and Respectively indicate time points and The inertial direction offset projection at the moment, Indicates the The disturbance gain coefficient output by the device attitude estimation module in a time period, represents the constant damping adjustment parameter; The dimension normalization process of parameters standardizes physical quantities into dimensionless form by introducing characteristic quantities, reference values ​​of angle offset, calibration scale of inertial projection, maximum amplitude of disturbance gain and dimensionless factor of damping parameter.

7. The method for intelligent shooting behavior prediction and optimization based on deep learning according to claim 1, characterized in that: The specific steps of S4 are: S401: Retrieving the structural change performance, the action impact performance, and the targeting trend label, arranging the labels in the same time period according to the same collection period, assigning numbers to the three label contents and aligning them, constructing a horizontal comparison matrix according to the collection time nodes, and generating a label corresponding sequence group; S402: Based on the tag-corresponding sequence groups, the expression types of the three tags in each group are compared to determine whether there are simultaneously deviated, disturbed, and stable tags of the same type. The cases with consistent type expression are recorded in the statistical table, the same-direction judgment value of the tags is calculated, and a classification number is generated to obtain the same-direction classification ratio of the tags; S403: Call the label same-direction classification ratio, filter the time period numbers where most labels belong to the same type, classify them into deviation, disturbance and stabilization trends according to type, mark and integrate the corresponding types of all time periods, and obtain the shooting behavior bias label.

8. The method for intelligent shooting behavior prediction and optimization based on deep learning according to claim 7, characterized in that: The calculation formula of the same-direction determination value of the tag is specifically as follows: ； in, Represents the same direction judgment value of the mth group of labels, Indicates the discrete difference of the mth group of deviation type labels, is the terrain penetration weight of the mth group of disturbance type labels, represents the baseline number of sample groups, Refers to the balance factor of the jth stabilizing label in the mth group, Characterizes the curvature change of the j-th label in the m-th group.

9. The method for intelligent shooting behavior prediction and optimization based on deep learning according to claim 1, characterized in that: The specific steps of S5 are: S501: calling the trend category in the shooting behavior bias tag, matching it to a pre-defined trend status prompt content table, extracting the index number set of prompt content corresponding to each trend category, filtering the related content range according to the tag type, and generating a trend prompt index mapping table; S502: extracting prompt item texts matching the tag numbers from the prompt content table according to the trend prompt index mapping table in a sequential order of the collection period, rearranging the prompt items in the order of the tag classification, constructing a structured prompt set, and obtaining a tag-pointing prompt group; S503: calling the tag to point to the prompt group, merging and sorting the prompt items according to the trend type, outputting the classified content in the order of deviation, disturbance and stabilization, generating a set of prediction and solution texts under the trend state, and obtaining the shooting behavior prediction and optimization results; The tag type filtering associated content range refers to limiting the call of all prompt item sets matching the tag category in the prompt content table according to the status classification corresponding to each trend tag; The prompt item text that matches the tag number refers to a text description set corresponding to the trend category number in the prompt content table, which is manually preset and associated according to the trend type coding index.

Images