Training data analysis method and system based on photoelectric shooting feedback

Abstract

The invention discloses a training data analysis method and system based on photoelectric shooting feedback. The method comprises the following steps of: a, acquiring multi-dimensional data; b, a data preprocessing step; c, an intelligent analysis step; d, a personalized feedback step; the system comprises a data acquisition unit (1) used for acquiring multi-source parameters in the shooting training process, a data processing unit (2) used for processing data, an intelligent analysis unit (3) used for supporting real-time data fusion and precision evaluation and defect diagnosis, and an intelligent control unit (4) used for 3D motion simulation and voice guidance. The interactive feedback unit (4) is used for outputting personalized training suggestions; and the training management unit (5) is used for establishing a trainee archive and supporting historical data tracing and progress trend analysis. The invention relates to the technical field of intelligent shooting training.

Description

Technical Field

[0001] This invention relates to the field of intelligent shooting training technology, and in particular to a training data analysis method and system based on photoelectric shooting feedback. Background Technology

[0002] Electro-optical shooting systems, with their advantages of safety, economy, and ease of deployment, have become the mainstream choice for athlete training, student training, and national defense education. However, current technology still has the following limitations: Single data dimension: Traditional systems only collect the location of the laser hitting the target or the time of the projectile's passage. They lack in-depth analysis of photoelectric signal characteristics (such as trajectory curvature and light intensity change rate) and shooting action process (such as gun holding stability and trigger pull timing), making it impossible to establish a "action-result" correlation model, resulting in unclear technical defects.

[0003] Delayed feedback mechanism: Most systems require a complete set of shooting (usually 10-30 shots) before batch analysis, and the feedback delay is generally >3 minutes. Students and novice athletes cannot correct their movements in time, which can easily lead to incorrect muscle memory and low training efficiency.

[0004] Lack of population fit: The use of a uniform assessment standard fails to differentiate the core needs of different training groups (e.g., students focus on movement standardization, while athletes focus on precision and stability). For example, student training overemphasizes accuracy, leading to insufficient training of basic postures; athlete training lacks targeted parameter optimization suggestions.

[0005] The algorithm model is simple: existing analyses mostly use statistical methods (such as calculating the average deviation), which cannot handle the nonlinear relationship between photoelectric signals and motion parameters. The accuracy of technical defect diagnosis is less than 65%, and the reference value is limited.

[0006] Existing patents, such as CN114543588A ("A Laser Shooting Training Evaluation System and Evaluation Method"), only evaluate the effectiveness of laser shooting training. CN114093030A ("A Shooting Training Analysis Method Based on Human Posture Learning") provides shooters with precise feedback during live-fire exercises and timely correction of errors, but it requires a large amount of ammunition and incurs high costs. Therefore, there is an urgent need to construct a low-cost, multi-dimensional, real-time feedback intelligent training analysis system that is adaptable to different user groups. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a training data analysis method and system based on photoelectric shooting feedback. Through multi-source data acquisition, deep model analysis and differentiated feedback, the method achieves "full-dimensional data coverage, real-time analysis and processing, and personalized guidance push" in shooting training, thereby solving the technical limitations of traditional systems, improving training efficiency and quality, and reducing costs.

[0008] The technical solution adopted in this invention is a training data analysis method based on photoelectric shooting feedback, which includes the following steps: a. Multi-dimensional data acquisition steps: The intensity change curve, trajectory coordinates (x, y, t), and beam spread angle of the shooting light signal are acquired through an optoelectronic sensor array; the gun-holding posture parameters and the force-time curve of the trigger pull are acquired through a 6-axis motion sensor; the hit position coordinates and bullet impact point distribution are obtained through a target positioning system; and the single-shot duration, continuous shooting interval, and aiming preparation time are recorded through a high-precision timer. b. Data preprocessing steps: Spatiotemporal synchronization of the collected data, noise removal using Kalman filtering, and extraction of multidimensional feature vectors; c. Intelligent analysis steps: Input the preprocessed data into a deep belief network model based on transfer learning. The deep belief network model is generated by training through several sets of labeled samples and outputs shooting accuracy indicators, technical defect labels and defect impact weights. d. Personalized feedback steps: Generate differentiated training suggestions based on the analysis results and trainee type, display them in real time through a visual interface, and dynamically adjust the training difficulty parameters based on the trainee's progress curve.

[0009] Furthermore, in step a, the gun-holding posture parameters include the pitch angle, yaw angle, and roll angle of the light-firing gun.

[0010] Furthermore, in step b, the multidimensional feature vector includes 5-dimensional photoelectric features, 6-dimensional motion features, 4-dimensional position features, and 5-dimensional time features. The photoelectric features include peak light intensity, beam spread angle, trajectory straightness deviation, signal rise time, and light speed. The motion features include pitch angle standard deviation, yaw angle standard deviation, peak trigger pull force, trigger pull time, aiming trajectory length, and attitude stabilization time. The position features include average bullet impact point x / y coordinates, bullet impact point distribution entropy, and maximum / minimum bullet impact point distance. The time features include single-shot firing duration, aiming preparation time, average firing interval, interval variation coefficient, and fastest firing reaction time.

[0011] Furthermore, in step c, the number of labeled samples is 5000 sets, including 2000 sets of student data and 3000 sets of athlete data. The shooting accuracy indicators include average radial deviation, bullet impact density, and hit rate. The technical defect labels include unstable posture, timing deviation, and aiming deviation.

[0012] Furthermore, in step d, the training difficulty parameters include target speed and shooting distance.

[0013] Furthermore, the deep belief network model is a CNN-LSTM hybrid model, which consists of an input layer, hidden layers, an output layer, and a transfer learning mechanism. The input layer takes in a multi-dimensional feature vector, and the hidden layer consists of four Restricted Boltzmann Machines (RBMs) with 128, 64, 32, and 16 neurons respectively. The output layer outputs three accuracy metrics, three defect labels, and five difficulty adjustment parameters. The transfer learning mechanism works as follows: during student group training, the weights of the first three hidden layers are frozen, and the output layer is fine-tuned; during athlete group training, all four hidden layer weights are unfrozen for adaptive learning.

[0014] A system for implementing the method described above, the system comprising A data acquisition unit for collecting multi-source parameters during shooting training includes a 12-channel laser photoelectric sensor, a 6-axis IMU sensor, a high-definition target camera, a piezoelectric force sensor, and a nanosecond-level timer. The data processing unit is used to process data. The data processing unit uses an FPGA chip for real-time noise reduction and feature extraction, providing high-quality input for the intelligent analysis unit. The device has the aforementioned deep belief network model, which is used to support intelligent analysis units for real-time data fusion, accuracy assessment, and defect diagnosis; An interactive feedback unit for 3D motion simulation and voice guidance, providing personalized training suggestions; and A training management unit used to establish a trainer profile database and support historical data tracing and progress trend analysis; The data acquisition unit, the data processing unit, the intelligent analysis unit, the interactive feedback unit, and the training management unit communicate via a high-speed bus.

[0015] Furthermore, the data acquisition unit also includes an environmental sensing module, which is used to dynamically correct the influence of environmental factors on photoelectric signals.

[0016] Furthermore, the interactive feedback unit supports an AR overlay module, which compares the real-time predicted trajectory of the impact point with the actual trajectory to help trainees understand and correct the direction.

[0017] The beneficial effects of this invention are as follows: This invention achieves "full-dimensional data coverage, real-time analysis, and personalized guidance" in shooting training through multi-source data acquisition, deep model analysis, and differentiated feedback. This overcomes the technical limitations of traditional systems, improves training efficiency and quality, and reduces costs. Compared to existing technologies, this invention has the following core advantages: Comprehensive data dimensions: By integrating 20-dimensional multi-source feature vectors, a correlation model of "photoelectric signal-action parameters-shooting result" is established, achieving a technical defect diagnosis accuracy of 93%, far exceeding that of traditional systems (<65%). Accuracy of population adaptation: Through transfer learning, differentiated analysis of students / athletes was achieved, resulting in a 58.9% improvement in students' movement standardization and a 35.2% improvement in athletes' precision stability. Real-time feedback efficiency: The entire process has a latency of ≤100ms, improving the timeliness of correcting students' incorrect movements by 80% and shortening the training achievement cycle by 42.9%; Wide range of applications: Supports laser / live-fire shooting and static / dynamic target training, suitable for schools, clubs, professional teams and other scenarios, and reduces deployment costs by 30% compared to traditional systems. Attached Figure Description

[0018] Figure 1 This is a simplified flowchart of the method of the present invention; Figure 2 This is a simplified block diagram of the system of the present invention. Detailed Implementation

[0019] like Figure 1 and Figure 2 As shown, this invention provides a training data analysis method based on photoelectric shooting feedback, which includes the following steps: a. Multi-dimensional data acquisition steps: The intensity change curve, trajectory coordinates (x, y, t), and beam spread angle of the shooting light signal are acquired through an optoelectronic sensor array; the gun-holding posture parameters and the force-time curve of the trigger pull are acquired through a 6-axis motion sensor; the hit position coordinates and bullet impact point distribution are obtained through a target positioning system; and the single-shot duration, continuous shooting interval, and aiming preparation time are recorded through a high-precision timer. b. Data preprocessing steps: Spatiotemporal synchronization of the collected data, noise removal using Kalman filtering, and extraction of multidimensional feature vectors; c. Intelligent analysis steps: Input the preprocessed data into a deep belief network model based on transfer learning. The deep belief network model is generated by training through several sets of labeled samples and outputs shooting accuracy indicators, technical defect labels and defect impact weights. d. Personalized feedback steps: Generate differentiated training suggestions based on the analysis results and trainee type, display them in real time through a visual interface, and dynamically adjust the training difficulty parameters based on the trainee's progress curve.

[0020] Furthermore, in step a, the gun-holding posture parameters include the pitch angle, yaw angle, and roll angle of the light-firing gun.

[0021] In step b, the number of multidimensional feature vectors is 20, including 5-dimensional photoelectric features, 6-dimensional motion features, 4-dimensional position features, and 5-dimensional time features. The photoelectric features include peak light intensity, beam spread angle, trajectory straightness deviation, signal rise time, and light speed. The motion features include pitch angle standard deviation, yaw angle standard deviation, peak trigger pull force, trigger pull time, aiming trajectory length, and attitude stabilization time. The position features include average x / y coordinates of the bullet impact point, bullet impact point distribution entropy, and maximum / minimum bullet impact point distance. The time features include single-shot firing duration, aiming preparation time, average firing interval, interval variation coefficient, and fastest firing reaction time.

[0022] In step c, the number of labeled samples is 5000 sets, including 2000 sets of student data and 3000 sets of athlete data. The shooting accuracy indicators include average radial deviation, bullet impact density and hit rate. The technical defect labels include unstable posture, timing deviation and aiming deviation.

[0023] In step d, the training difficulty parameters include target speed and shooting distance.

[0024] The deep belief network model is a CNN-LSTM hybrid model, which consists of an input layer, hidden layers, an output layer, and a transfer learning mechanism. The input layer takes in a multi-dimensional feature vector. The hidden layer consists of four Restricted Boltzmann Machines (RBMs), with 128, 64, 32, and 16 neurons respectively. The output layer outputs three accuracy metrics, three defect labels, and five difficulty adjustment parameters. The transfer learning mechanism works as follows: during student group training, the weights of the first three hidden layers are frozen, and the output layer is fine-tuned; during athlete group training, all four hidden layer weights are unfrozen for adaptive learning.

[0025] A system for implementing the method described above, the system comprising Data acquisition unit 1 is used to collect multi-source parameters during shooting training. The data acquisition unit 1 includes 12 laser photoelectric sensors, a 6-axis IMU sensor, a target surface high-definition camera, a piezoelectric force sensor and a nanosecond timer. The data processing unit 2 is used to process data. The data processing unit 2 uses an FPGA chip for real-time noise reduction and feature extraction, providing high-quality input for the intelligent analysis unit 3. The device includes the deep belief network model, which is used to support intelligent analysis unit 3 for real-time data fusion, accuracy assessment, and defect diagnosis; Interactive feedback unit 4 for 3D motion simulation and voice guidance, outputting personalized training suggestions; and Training management unit 5, used to establish a trainer profile database and support historical data tracing and progress trend analysis; The data acquisition unit 1, the data processing unit 2, the intelligent analysis unit 3, the interactive feedback unit 4, and the training management unit 5 communicate via a high-speed bus.

[0026] More specifically, the data acquisition unit 1 further includes an environmental perception module (temperature -20~60℃±0.3℃, humidity 10~90%RH±2%, wind speed 0~20m / s±0.1m / s), which is used to dynamically correct the influence of environmental factors on photoelectric signals. The interactive feedback unit 4 supports an AR overlay module, which compares the real-time predicted trajectory of the impact point with the actual trajectory to help the trainee understand the correction direction.

[0027] The present invention will now be described in more detail.

[0028] The process of the method of the present invention is as follows.

[0029] (I) Multi-dimensional data collection Photoelectric signal acquisition: Four monitoring planes (3m, 6m, 9m, and 10m from the target) are set up along the firing path. Each plane is equipped with 12 laser sensors to collect the light intensity changes (sampling interval 50μs), trajectory coordinates (absolute position on the x / y axis), and diffusion angle (reflecting the stability of the projectile) as the beam passes through. Action parameter acquisition: A 6-axis IMU sensor is installed on the gun body of the photoelectric transmitter to record the gun's pitch angle (range -3° to +3°), yaw angle (range -3° to +3°), and angle change rate in real time; a piezoelectric sensor is installed at the trigger to collect the trigger pull force (0 to 50N) and force-time curve (to identify erroneous actions such as "slamming"). Position and time acquisition: The target surface 4K camera identifies the bullet impact point coordinates (accuracy ±0.2mm) and calculates the bullet impact point distribution entropy (reflecting density); the nanosecond-level timer records the preparation time from aiming to firing, the duration of a single shot, and the interval between consecutive shots.

[0030] (ii) Data preprocessing Spatiotemporal synchronization: Based on the photoelectric triggering time of the third monitoring plane, the action, position and time data are aligned, and the synchronization error is controlled within 0.5ms; Noise filtering: The photoelectric signal and motion data are processed using the Kalman filter algorithm. After filtering, the signal-to-noise ratio is improved from 25dB to over 40dB, and the angle measurement error is reduced by 60%. Feature extraction: Calculate a 20-dimensional feature vector from the original data to form structured input data.

[0031] (III) Intelligent Analysis and Modeling Model architecture: A CNN-BiLSTM hybrid model is used as a deep belief network model to construct nonlinear mapping relationships. Feature dimensionality reduction and association learning are performed through a 4-layer restricted Boltzmann machine to solve the complex coupling problem between photoelectric signals and action parameters. Transfer learning: The basic model is pre-trained using 5,000 sets of historical data, and only 50 sets of data are needed to adapt it for new trainees (adaptation time ≤ 2 minutes for students and ≤ 5 minutes for athletes). Output results include accuracy indicators (average radial deviation, 90% impact point density radius, etc.), technical defect labels (such as "excessive pitch angle fluctuation"), and defect impact weights (quantifying the degree of influence of each factor on accuracy).

[0032] (iv) Personalized feedback optimization Student training: Focus on providing suggestions for proper action, such as "If the yaw angle of holding the gun exceeds 1.2°, it is recommended to adjust the grip posture and perform 15 sets of static gun-holding training every day (45 seconds per set)", and demonstrate the correct action through AR animation; Athlete training: Focus on precision optimization parameters, such as "When the aiming preparation time is <1.5s, the hit rate drops by 23%, and it is recommended to extend the preparation time to 1.8-2.2s", and provide wind disturbance compensation at the same time; Dynamic adjustment: The training difficulty is automatically increased based on the trainee's progress rate (the deviation decreases by more than 15% in 3 consecutive training sessions). For example, the speed of the moving target is increased from 1m / s to 1.5m / s.

[0033] Compared with existing technologies, this invention achieves, for the first time, deep coupling analysis of photoelectric signal characteristics (trajectory spread angle, light intensity change rate) with action, position, and time data, achieving a technical defect diagnosis accuracy of 93%, a 43% improvement over traditional systems. Through transfer learning, the model is rapidly adapted to students / athletes, improving student group training movement standardization scores by 46% and athlete precision stability by 35%. With a latency of ≤100ms from data acquisition to suggestion output, the timeliness of correcting student errors is improved by 80%, preventing the formation of erroneous muscle memory. Based on the progress curve, training difficulty is automatically adjusted, shortening the student achievement period by 40% and increasing athlete specific skill improvement efficiency by 30%.

[0034] To verify the feasibility, advancement, and superiority of this invention, two controlled experiments were designed: basic training for students and specialized training for athletes. The experimental data are as follows: Example 1: Basic shooting training for college students (10m laser pistol) 1. Experimental Environment Training participants: 40 second-year university students (no shooting experience, 20 males and 20 females), randomly divided into an experimental group (the system of this invention) and a control group (traditional photoelectric target system), with 20 people in each group; Training equipment: 10m laser pistol (modified with the acquisition unit of this invention), fixed target (150mm in diameter, 15mm in core diameter); Training objective: Master basic aiming and firing techniques; achieve a hit rate of ≥50% with 100 training rounds. Training cycle: 3 times a week, 60 minutes each time, for 4 consecutive weeks.

[0035] 2. Data Acquisition and Processing Photoelectric data: Initial laser beam diffusion angle 2.5°±0.8°, trajectory straightness deviation 8.2±3.5mm, after filtering the diffusion angle measurement error reduced to 0.3°; Action data: initial pitch angle standard deviation 1.6°±0.5°, yaw angle standard deviation 1.4°±0.4°, trigger pull force fluctuation 4.2±1.8N; Position data: initial average radial deviation 12.5±4.2mm, impact point distribution entropy 0.85±0.12; Time data: initial aiming preparation time 1.2±0.5s, firing interval coefficient of variation 35%±10%.

[0036] 3. Intelligent Analysis and Feedback Defect diagnosis: The model identified the core defects as "unstable gun-holding posture (affecting weight 65%)" and "premature trigger pull (affecting weight 25%)". Personalized recommendations: "Add two sets of posture stability training sessions daily (10 minutes per set), requiring pitch angle fluctuations ≤0.8°; add a 0.5s stabilization period before pulling the trigger, and control the force at 22-25N." Feedback format: The AR overlay displays the deviation between the ideal gun-holding angle and the actual angle in real time (accuracy ±0.1°), with a voice prompt "0.6° to the left, please fine-tune your right arm".

[0037] 4. Comparison of training effects

[0038] Conclusion: The experimental group showed significantly better performance than the control group in terms of basic movement standardization and accuracy, verifying the suitability and effectiveness of the invention for students.

[0039] Example 2: Rifle-specific training for shooting athletes (50m rifle) 1. Experimental Environment Training participants: 8 national level 2 or above rifle athletes (3-8 years of experience in the sport, 5 males and 3 females); Training equipment: 50m electro-optical rifle (modified with data acquisition unit), movable target (speed 0-3m / s); Training objective: Improve dynamic shooting accuracy, with 90% of the bullet impact points having a radius (R90) ≤ 8mm within 10 rounds / 60s; Training cycle: 5 times a week, 120 minutes each time, for 6 consecutive weeks.

[0040] 2. Data Acquisition and Processing Photoelectric data: The trajectory of the projectile deviated by 1.2±0.5mm due to wind speed (wind speed 1~2m / s), and the rate of change of light intensity was strongly correlated with stability (correlation coefficient 0.82). Action data: During dynamic firing, the pitch angle fluctuates by 0.7±0.2° (static 0.3±0.1°), and the trigger pull time is extended to 0.6±0.1s; Position data: Initial R90 = 10.5 ± 1.2 mm (static), 13.8 ± 1.5 mm (dynamic); Time data: The coefficient of variation for firing interval is 18% ± 5%, and the hit rate decreases by 22% when the preparation time is < 1.5s.

[0041] 3. Intelligent Analysis and Feedback Defect diagnosis: The model identified "a 31% increase in trajectory deviation during dynamic shooting (weight 48%)" and "a decrease in action stability during rapid shooting (weight 32%)". Personalized suggestions: "When the wind speed is v (m / s), the aiming point offset = 0.4v + 0.2mm; when the dynamic target speed is > 2m / s, the preparation time is extended to 1.8~2.0s", and wind disturbance compensation training programs are pushed out simultaneously. Feedback format: The touch screen displays the correlation curve of "trajectory offset - wind speed - motion", quantifying the best correction parameters under different wind speeds.

[0042] 4. Comparison of training results (System of this invention vs. traditional training)

[0043] Conclusion: This invention can significantly improve athletes' shooting accuracy and stability, and shorten the training cycle by 33.3%, verifying its advanced nature in professional training.

[0044] This invention collects the trajectory features of light signals during shooting using a distributed photoelectric sensor array, simultaneously acquiring shooting action parameters, hit position coordinates, and time-series data. After multi-source data fusion and feature extraction, an improved deep belief network model is used for real-time analysis to generate shooting accuracy assessment results, technical defect diagnosis reports, and personalized training plans. This invention solves the problems of single data dimension, delayed feedback, and insufficient targeted suggestions in traditional photoelectric shooting systems. Experimental data shows that it can improve the training efficiency of student groups by more than 40% and the shooting accuracy stability of athletes by 35%, making it suitable for shooting training scenarios at all levels.

[0045] Finally, it should be emphasized that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A training data analysis method based on photoelectric shooting feedback, characterized in that, The method includes the following steps: a. Multi-dimensional data acquisition steps: The intensity change curve, trajectory coordinates (x, y, t), and beam spread angle of the shooting light signal are acquired through an optoelectronic sensor array; the gun-holding posture parameters and the force-time curve of the trigger pull are acquired through a 6-axis motion sensor; the hit position coordinates and bullet impact point distribution are obtained through a target positioning system; and the single-shot duration, continuous shooting interval, and aiming preparation time are recorded through a high-precision timer. b. Data preprocessing steps: Spatiotemporal synchronization of the collected data, noise removal using Kalman filtering, and extraction of multidimensional feature vectors; c. Intelligent analysis steps: Input the preprocessed data into a deep belief network model based on transfer learning. The deep belief network model is generated by training through several sets of labeled samples and outputs shooting accuracy indicators, technical defect labels and defect impact weights. d. Personalized feedback steps: Generate differentiated training suggestions based on the analysis results and trainee type, display them in real time through a visual interface, and dynamically adjust the training difficulty parameters based on the trainee's progress curve.

2. The training data analysis method based on photoelectric shooting feedback according to claim 1, characterized in that, In step a, the gun-holding posture parameters include the pitch angle, yaw angle, and roll angle of the light-firing gun.

3. The training data analysis method based on photoelectric shooting feedback according to claim 1, characterized in that, In step b, the multidimensional feature vector includes 5-dimensional photoelectric features, 6-dimensional motion features, 4-dimensional position features, and 5-dimensional time features. The photoelectric features include peak light intensity, beam spread angle, trajectory straightness deviation, signal rise time, and light speed. The motion features include pitch angle standard deviation, yaw angle standard deviation, peak trigger pull force, trigger pull time, aiming trajectory length, and attitude stabilization time. The position features include average bullet impact point x / y coordinates, bullet impact point distribution entropy, and maximum / minimum bullet impact point distance. The time features include single-shot firing duration, aiming preparation time, average firing interval, interval variation coefficient, and fastest firing reaction time.

4. The training data analysis method based on photoelectric shooting feedback according to claim 1, characterized in that, In step c, the number of labeled samples is 5000 sets, including 2000 sets of student data and 3000 sets of athlete data. The shooting accuracy indicators include average radial deviation, bullet impact density and hit rate. The technical defect labels include unstable posture, timing deviation and aiming deviation.

5. The training data analysis method based on photoelectric shooting feedback according to claim 1, characterized in that, In step d, the training difficulty parameters include target speed and shooting distance.

6. A training data analysis method based on photoelectric shooting feedback according to claim 1, characterized in that, The deep belief network model is a CNN-LSTM hybrid model, which consists of an input layer, hidden layers, an output layer, and a transfer learning mechanism. The input layer takes in a multi-dimensional feature vector. The hidden layer consists of four Restricted Boltzmann Machines (RBMs), with 128, 64, 32, and 16 neurons respectively. The output layer outputs three accuracy metrics, three defect labels, and five difficulty adjustment parameters. The transfer learning mechanism works as follows: during student group training, the weights of the first three hidden layers are frozen, and the output layer is fine-tuned; during athlete group training, all four hidden layer weights are unfrozen for adaptive learning.

7. A system for implementing the method as described in any one of claims 1 to 6, characterized in that, The system includes The data acquisition unit (1) is used to collect multi-source parameters during shooting training. The data acquisition unit (1) includes a 12-channel laser photoelectric sensor, a 6-axis IMU sensor, a target surface high-definition camera, a piezoelectric force sensor and a nanosecond timer. The data processing unit (2) is used to process the data. The data processing unit (2) uses an FPGA chip for real-time noise reduction and feature extraction to provide high-quality input for the intelligent analysis unit (3). The device has the aforementioned deep belief network model, which is used to support intelligent analysis units (3) for real-time data fusion, accuracy assessment, and defect diagnosis. An interactive feedback unit (4) for 3D motion simulation and voice guidance, outputting personalized training suggestions; and (5) Training management unit used to establish a trainer profile database to support historical data tracing and progress trend analysis. The data acquisition unit (1), the data processing unit (2), the intelligent analysis unit (3), the interactive feedback unit (4), and the training management unit (5) communicate with each other via a high-speed bus.

8. The system according to claim 7, characterized in that, The data acquisition unit (1) also includes an environmental sensing module, which is used to dynamically correct the influence of environmental factors on photoelectric signals.

9. The system according to claim 7, characterized in that, The interactive feedback unit (4) supports an AR overlay module, which compares the real-time predicted trajectory of the impact point with the actual trajectory to help trainees understand and correct the direction.

Images