An unmanned aerial vehicle flight data acquisition intelligent control system

Abstract

The application provides a kind of unmanned plane flight data acquisition intelligent control system, belong to data processing technical field, including: data acquisition module, offset stationary degree calculation module, confidence acquisition module and acquisition control module, according to the flight state of unmanned plane deviates stationary degree, and the difference between historical flight attitude data and its predicted value, obtain the attitude prediction failure degree of unmanned plane at current time;According to the attitude prediction failure degree of unmanned plane, obtain the prediction window confidence attenuation coefficient of LSTM model at current time;Utilize prediction window confidence attenuation coefficient to actively compress candidate window, compare the effective prediction window length after compression with the minimum exposure time of camera bottom layer physically limit, if unable to meet the exposure requirement, then forced to lock acquisition function;The application improves the data acquisition quality of unmanned plane in high dynamic environment.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to an intelligent control system for acquiring flight data from unmanned aerial vehicles (UAVs). Background Technology

[0002] With the widespread application of drone technology in fields such as geographic surveying, power line inspection, and security monitoring, the requirements for the quality of flight data (especially image and video data) are increasing. To obtain high-quality flight data, drones need to maintain stable attitude during flight to avoid image blurring caused by aircraft vibration. However, the actual operating environment of drones is usually quite complex, often involving high-speed flight or exposure to high-dynamic environments such as gusts and air shear.

[0003] In existing technologies, to address the impact of flight attitude on data acquisition quality, a time-series prediction method based on Long Short-Term Memory (LSTM) networks is typically employed. This method leverages the powerful sequence learning capabilities of LSTM to predict future flight attitude changes by analyzing historical data, thereby identifying a relatively stable acquisition window for data collection.

[0004] However, when encountering sudden strong airflow or needing to quickly avoid obstacles, LSTM may experience a "memory failure" problem. Since LSTM-based prediction methods mainly rely on past data patterns for prediction, when the UAV encounters strong gusts or makes sudden maneuvers, the changes in flight attitude become very chaotic. At this time, historical data cannot provide effective reference and may instead cause the model to make lagging or incorrect judgments. This "memory failure" problem will cause the UAV to incorrectly execute data acquisition commands when its attitude is most unstable, or miss the best acquisition opportunity, thus seriously affecting the quality and validity of the data. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an intelligent control system for unmanned aerial vehicle (UAV) flight data acquisition.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The data acquisition module is used to acquire the historical flight attitude data sequence of the UAV, as well as the time required for a single exposure of the camera carried by the UAV. The offset stability calculation module is used to extract angular velocity data from the historical flight attitude data sequence, calculate the degree of sudden change in the force on the UAV fuselage and the discrete volatility of the flight status data at the current moment; and combine the degree of sudden change in the force with the discrete volatility of the flight status data to obtain the deviation stability of the UAV's flight status at the current moment. The confidence acquisition module is used to input the historical flight attitude data sequence into the LSTM model to obtain the predicted flight attitude data sequence of the UAV in the future time period; based on the difference between the predicted flight attitude data and the actual flight attitude data of the UAV at the current moment, combined with the flight state deviation stability, to obtain the attitude prediction failure degree of the UAV at the current moment; and based on the attitude prediction failure degree, to obtain the prediction window confidence decay coefficient of the LSTM model at the current moment. The acquisition and control module is used to identify continuous time periods in the predicted flight attitude data sequence where attitude fluctuations meet preset static conditions as candidate prediction window lengths; dynamically correct and compress the candidate prediction window lengths using the prediction window confidence decay coefficient to obtain an effective prediction window length; if the effective prediction window length is greater than or equal to the time required for a minimum exposure of the UAV carrying the camera, the UAV is controlled to perform image acquisition within the corresponding effective prediction window; if the effective prediction window length is less than the time required for a minimum exposure of the UAV carrying the camera, it is determined that the current period is a pseudo-stationary period of LSTM model memory failure, and the image acquisition function of the UAV is locked.

[0007] Preferably, the calculation of the sudden change in stress on the UAV fuselage and the discrete volatility of flight status data at the current moment includes: Preset a time parameter The closest time before the current moment A time window consisting of several moments is denoted as the backtracking time window of the current moment; Obtain the drone's position at all times within the backtracking time window. The standard deviation of the angular velocity data along the axis is denoted as the value of the UAV at the current moment. Discrete volatility of flight state data along the axial direction; The drone at the current moment The angular velocity data along the axis is compared with the previous moment when the UAV was The difference between the angular velocity data along the axis is denoted as the value of the angular velocity of the UAV at the current moment. The angular velocity difference along the axis; the absolute value of the difference between the current angular velocity difference and the previous angular velocity difference is denoted as the value of the difference between the UAV body and the angular velocity difference along the axis at the current moment. The degree of abrupt change in force along the axial direction.

[0008] Preferably, the step of combining the severity of the sudden force change with the discrete volatility of the flight state data to obtain the deviation of the UAV's flight state from stability at the current moment includes: The drone at the current moment Discrete fluctuations in flight state data along the axial direction and the current position of the UAV fuselage in... The product of the abrupt changes in force along the axial direction is denoted as the stress on the UAV fuselage at the current moment. The flight state in the axial direction deviates from the stability factor; The flight state deviation from stability of the UAV at the current moment is obtained by summing the deviation factors of the UAV's flight state in the three axes at the current moment.

[0009] Preferably, obtaining the attitude prediction failure level of the UAV at the current moment includes: The predicted flight attitude data sequence includes predicted angular velocity data and predicted velocity data of the UAV in three axes. Based on the difference between the predicted flight attitude data and the actual flight attitude data of the UAV at the current moment, obtain the attitude difference value of the UAV at the current moment. Based on the attitude difference value at each moment in the backtracking time window, obtain the attitude prediction failure factor at each moment; By combining the attitude prediction failure factors at all moments in the backtracking time window with the flight state deviation stability, the degree of attitude prediction failure of the UAV at the current moment can be obtained.

[0010] Preferably, obtaining the attitude difference value of the UAV at the current moment includes: The drone at the current moment The absolute value of the difference between the actual angular velocity data and the predicted angular velocity data in the axial direction is denoted as the angular velocity difference. The drone at the current moment The absolute value of the difference between the actual velocity data and the predicted velocity data along the axis is denoted as the velocity difference. The product of the velocity difference and the angular velocity difference is denoted as the value of the UAV's velocity at the current moment. Attitude difference factor along the axis; The summation and normalization of the attitude difference factors in the three axes of the UAV at the current moment are recorded as the attitude difference value of the UAV at the current moment.

[0011] Preferably, obtaining the attitude prediction failure factor at each time step includes: The number of all moments in the backtracking time window is compared with the number of moments in the backtracking time window. The inverse proportional value of the difference between the index values ​​corresponding to the nth time point is denoted as the nth time point. Time weight of each moment; The first The time weight of the i-th moment and the i-th moment The product of the attitude difference values ​​of the UAV at time i is denoted as the i-th time. Attitude prediction failure factor at each time step.

[0012] Preferably, the step of combining the attitude prediction failure factors at all times within the backtracking time window with the flight state deviation stability to obtain the attitude prediction failure level of the UAV at the current time includes: The normalized value of the product of the mean of the attitude prediction failure factors at all times in the backtracking time window at the current time and the deviation of the UAV's flight state from stability at the current time is taken as the attitude prediction failure degree of the UAV at the current time.

[0013] Preferably, obtaining the prediction window confidence decay coefficient of the LSTM model at the current time based on the attitude prediction failure degree includes: Preset a model reliability threshold and a model failure threshold If the attitude prediction failure level of the UAV at the current moment is less than or equal to the model reliability threshold. Then This serves as the confidence decay coefficient for the prediction window of the LSTM model at the current moment; if the attitude prediction failure level of the UAV at the current moment is greater than the model reliability threshold... And less than the model failure threshold The attitude prediction failure level is directly used as the confidence decay coefficient of the prediction window of the LSTM model at the current moment; if the attitude prediction failure level of the UAV at the current moment is greater than or equal to the model failure threshold... Then This serves as the confidence decay coefficient for the prediction window of the LSTM model at the current moment.

[0014] Preferably, the step of dynamically correcting and compressing the candidate prediction window length using the prediction window confidence decay coefficient to obtain the effective prediction window length includes: The final effective prediction window length is the floor value of the product between the candidate prediction window length and the prediction window confidence decay coefficient of the LSTM model at the current time.

[0015] Preferably, the step of identifying the continuous time period during which attitude fluctuations satisfy a preset static condition as the candidate prediction window length includes: Calculate the change in attitude angular velocity between adjacent prediction times in the predicted flight attitude data sequence; identify the time period in which the change in attitude angular velocity is continuously lower than a preset fluctuation safety threshold as a candidate prediction window where the attitude is relatively still, and record the time span of this period as the length of the candidate prediction window.

[0016] The intelligent control system for UAV flight data acquisition provided by this invention has the following beneficial effects: This invention combines the intensity of sudden force changes at the physical level with the discrete volatility of flight state data at the statistical level to construct the flight state deviation stability of UAVs, which can detect disordered fluctuations such as sudden airflow or abrupt maneuvers in the environment before the LSTM model. On this basis, the physical fluctuation feature is combined with the historical prediction residuals of the LSTM model to quantify the degree of attitude prediction failure at the current moment and map it into the confidence decay coefficient of the prediction window. This allows the system to move away from blindly relying on the LSTM output window and instead search for relatively stationary candidate windows within the prediction sequence. Then, before the LSTM model experiences memory failure and a pseudo-stationary period, the system actively compresses these candidate windows using a decay coefficient. The length of the compressed effective prediction window is then compared to the physical limit of the camera's minimum exposure time. If the exposure requirement cannot be met, the acquisition function is forcibly locked. This complete closed-loop control strategy effectively avoids motion blur-prone images caused by model prediction lag, thus significantly improving the data acquisition quality of UAVs in highly dynamic environments. Attached Figure Description

[0017] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the steps of an intelligent control system for acquiring flight data of an unmanned aerial vehicle (UAV) according to an exemplary embodiment of the present invention. Figure 2 This is a flight state deviation stability change curve of an intelligent control system for UAV flight data acquisition provided by the present invention according to an exemplary embodiment; Figure 3 This is a comparison diagram of data acquisition execution strategies of an intelligent control system for UAV flight data acquisition provided by the present invention according to an exemplary embodiment. Detailed Implementation

[0019] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0020] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] First, this invention provides an intelligent control system for acquiring flight data from unmanned aerial vehicles (UAVs), specifically as follows: Figure 1 As shown, it includes the following steps: The data acquisition module is used to acquire the historical flight attitude data sequence of the UAV, as well as the time required for a single exposure of the camera carried by the UAV.

[0022] The system reads the UAV's flight attitude data at each moment in the historical time period through the high-frequency IMU sensor interface of the UAV flight control system. The flight attitude data includes three-axis acceleration data and three-axis angular velocity data.

[0023] For the UAV flight attitude data collected over a historical period, the system performs a low-pass filtering algorithm to remove high-frequency vibration noise and uses a maximum-minimum normalization method to map the data to a standard interval. Subsequently, the UAV flight attitude data is strictly time-series aligned based on timestamps to obtain the UAV's historical flight attitude data sequence.

[0024] Obtain the minimum exposure time required for a single exposure by referring to the camera specifications on the drone.

[0025] The low-pass filtering algorithm and the maximum-minimum normalization method are existing technologies, and will not be described in detail here.

[0026] At this point, the historical flight attitude data sequence of the UAV was obtained.

[0027] The offset stability calculation module is used to extract angular velocity data from the historical flight attitude data sequence, calculate the degree of sudden change in the force on the UAV body and the discrete volatility of the flight state data at the current moment; and combine the degree of sudden change in the force with the discrete volatility of the flight state data to obtain the deviation stability of the UAV's flight state at the current moment.

[0028] It should be noted that existing LSTM models, when processing flight attitude prediction, are essentially based on the assumption of continuity in flight attitude data, meaning they assume that the future state distribution is correlated with historical data. However, when a UAV encounters sudden airflow impacts or performs abrupt maneuvers, its most obvious characteristic is an abnormal change in "acceleration." This abrupt change usually occurs before changes in velocity or position, so relying solely on the first or second derivatives of (angular velocity and angular acceleration) is insufficient to effectively distinguish between normal, smooth flight and sudden turbulent disturbances.

[0029] It should be further noted that during normal flight, the slight adjustments made by the UAV during turns, climbs, and hovering will cause significant changes in angular velocity. However, these changes are smooth transitions, and the changes in angular acceleration are relatively gentle and continuous, but lack suddenness. Therefore, it is necessary to construct the deviation of the UAV's flight state from the stable state at the current moment. This is done by analyzing the changes in flight attitude data within the historical flight attitude data sequence to quantify the degree of deviation between the current flight state and the stable state of the UAV, thereby more accurately judging the flight state.

[0030] Preferably, in one embodiment of the present invention, the specific method for obtaining the deviation of the UAV's flight state from stability at the current moment by analyzing the changes in flight attitude data within a historical flight attitude data sequence is as follows: Preset a time parameter In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. It depends on the specific implementation situation; The nearest time between the current moments A time window consisting of several moments is denoted as the backtracking time window of the current moment; The standard deviation of the angular velocity data of the UAV in the k-axis direction at all times in the backtracking time window at the current time is denoted as the discrete fluctuation of the UAV's flight state data in the k-axis direction at the current time. The difference between the angular velocity data of the drone in the k-axis direction at the current moment and the angular velocity data of the drone in the k-axis direction at the previous moment is recorded as the angular velocity difference value of the drone in the k-axis direction at the current moment; the absolute value of the difference between the angular velocity difference value of the drone in the k-axis direction at the current moment and the angular velocity difference value of the drone in the k-axis direction at the previous moment is recorded as the degree of sudden change in force on the drone body in the k-axis direction at the current moment. The product of the discrete volatility of the UAV's flight state data in the k-axis direction at the current moment and the severity of the sudden change in force on the UAV's body in the k-axis direction at the current moment is denoted as the flight state deviation from stability factor of the UAV's body in the k-axis direction at the current moment. The flight state deviation from stability of the UAV at the current moment is obtained by summing the flight state deviation from stability factors of the UAV's body in the three axes at the current moment. The specific formula is as follows: In the formula, This indicates the deviation of the drone's flight state from stability at the current moment; This indicates the degree of sudden change in force on the drone's body along the k-axis at the current moment; This represents the standard deviation of the angular velocity data of the UAV in the k-axis direction across all time points within the backtracking time window at the current moment; Indicates taking the absolute value; This represents the logarithmic function with the natural constant as the base.

[0031] It should be noted that as the intensity of the sudden external airflow impact increases, the aerodynamic torque experienced by the UAV undergoes a sudden and drastic change, leading to a significant increase in the severity of the force abrupt change. Simultaneously, the randomness of the disturbance increases the discrete volatility of the flight state data, causing the cumulative term within the brackets to grow non-linearly, meaning the UAV's flight state deviates rapidly from stability. This process combines the physical "torque abrupt change" with the statistical "discrete distribution," enabling it to detect the environment entering a state of disorder and fluctuation before the LSTM model, thereby triggering a reassessment of the reliability of the prediction model.

[0032] Please see Figure 2 The figure shows the flight state deviation from stability curve of an intelligent control system for UAV flight data acquisition. During the stable flight period of 0-400ms, the flight state deviation from stability is almost 0. However, at the moment the disturbance begins at 400ms, the curve spikes up instantly. This proves that the present invention can capture the starting point of disordered fluctuations more sensitively than the LSTM model.

[0033] At this point, the deviation of the drone's flight status from stability at the current moment is obtained.

[0034] The confidence acquisition module is used to input the historical flight attitude data sequence into the LSTM model to obtain the predicted flight attitude data sequence of the UAV in the future time period; based on the difference between the predicted flight attitude data and the actual flight attitude data of the UAV at the current moment, combined with the flight state deviation stability, to obtain the attitude prediction failure degree of the UAV at the current moment; and based on the attitude prediction failure degree, to obtain the prediction window confidence decay coefficient of the LSTM model at the current moment.

[0035] It's important to note that the prediction failure of LSTM models manifests as a significant gap between "historical experience" and "current reality" during algorithm execution. When the environment changes drastically, the parameter weights learned from long historical sequences cannot adapt to the current input distribution, leading to a sharp increase in the error between the predicted and true values. Common interfering factors in this process include single-point prediction bias caused by sporadic sensor noise. This bias is usually transient and random, and quickly recovers to normal range in the next time step. Existing techniques typically rely solely on the current instantaneous error to determine model usability or simply detect whether the external environment is harsh. Their limitations are: instantaneous errors are easily affected by noise, leading to misjudgments (overly sensitive); while simply detecting the environment ignores the model's own robustness (overly insensitive).

[0036] To address the aforementioned issues, this embodiment introduces "the degree of failure in UAV attitude prediction" and "the confidence decay coefficient of the prediction window." These two indicators, by associating physical change characteristics with historical prediction residuals, enable dynamic correction of the weights of the LSTM output, thereby improving the model's adaptability and reliability in complex environments.

[0037] Preferably, in one embodiment of the present invention, the specific method for obtaining the degree of attitude prediction failure of the UAV at the current moment based on the deviation of the UAV's flight state from stability and the difference between historical flight attitude data and its predicted value is as follows: The historical flight attitude data sequence of the UAV is input into the LSTM model to obtain the predicted flight attitude data of the UAV at the current moment; the predicted flight attitude data of the UAV at the current moment includes the predicted angular velocity data and predicted velocity data of the UAV in the three-axis directions; The absolute value of the difference between the current angular velocity data of the UAV in the k-axis direction and the predicted angular velocity data is denoted as the angular velocity difference of the UAV in the k-axis direction at the current moment; the absolute value of the difference between the current velocity data of the UAV in the k-axis direction and the predicted velocity data is denoted as the velocity difference of the UAV in the k-axis direction at the current moment; the product of the velocity difference of the UAV in the k-axis direction and the angular velocity difference is denoted as the attitude difference factor of the UAV in the k-axis direction at the current moment. The summation and normalization of the attitude difference factors of the UAV in the three axes at the current moment are recorded as the attitude difference value of the UAV at the current moment. The inverse proportional value between the number of all times in the backtrack time window at the current time and the sequence number value corresponding to the t-th time in the backtrack time window at the current time is denoted as the time weight of the t-th time. The product of the time weight of the t-th time in the backtracking time window of the current time and the attitude difference value of the UAV at the t-th time is denoted as the attitude prediction failure factor at the t-th time. The normalized value of the product of the mean of the attitude prediction failure factors at all times in the backtracking time window at the current time and the deviation of the UAV's flight state from stability at the current time is taken as the attitude prediction failure degree of the UAV at the current time. The specific formula is as follows: In the formula, This indicates the degree of failure in the attitude prediction of the UAV at the current moment; This indicates the deviation of the drone's flight state from stability at the current moment; This indicates the number of times within the backtracking time window at the current moment; This represents the attitude prediction failure factor at time t within the backtracking time window of the current moment; Represents an exponential function with the natural constant as its base; This represents the linear normalization function.

[0038] Preferably, in one embodiment of the present invention, the specific method for obtaining the prediction window confidence decay coefficient of the LSTM model at the current moment based on the degree of attitude prediction failure of the UAV is as follows: Preset a model reliability threshold and a model failure threshold In this embodiment, and This example is used for illustration; no specific limitations are set in this embodiment. and It depends on the specific implementation situation; If the attitude prediction failure level of the UAV at the current moment is less than or equal to the model reliability threshold If 1 is used as the confidence decay coefficient of the prediction window of the LSTM model at the current time, then if the attitude prediction failure of the UAV at the current time is greater than the model reliability threshold... And less than the model failure threshold The degree of attitude prediction failure of the UAV at the current moment is used as the confidence decay coefficient of the prediction window of the LSTM model at the current moment; if the confidence decay coefficient of the prediction window of the LSTM model at the current moment is greater than or equal to the model failure threshold... If 0 is used as the confidence decay coefficient of the prediction window of the LSTM model at the current time, then 0 will be used as the confidence decay coefficient of the prediction window at the current time.

[0039] It should be noted that as the deviation of flight state from stability increases (due to intensified physical disturbances) and the attitude prediction failure factor (historical prediction residuals) increases (due to a decline in historical model performance), the degree of attitude prediction failure of the UAV increases. When the attitude prediction failure of the UAV exceeds the safety threshold... Subsequently, the prediction window confidence decay coefficient begins to decrease. This process mathematically realizes the mapping from "physical perception" to "algorithm confidence". As environmental instability intensifies, the prediction window confidence decay coefficient becomes 0, thereby triggering the downgrade processing of the prediction results.

[0040] The LSTM model is an existing technology, and will not be described in detail here.

[0041] Thus, the confidence decay coefficient of the prediction window of the LSTM model at the current time is obtained.

[0042] The acquisition and control module is used to identify continuous time periods in the predicted flight attitude data sequence where attitude fluctuations meet preset static conditions as candidate prediction window lengths; dynamically correct and compress the candidate prediction window lengths using the prediction window confidence decay coefficient to obtain an effective prediction window length; if the effective prediction window length is greater than or equal to the time required for a minimum exposure of the UAV carrying the camera, the UAV is controlled to perform image acquisition within the corresponding effective prediction window; if the effective prediction window length is less than the time required for a minimum exposure of the UAV carrying the camera, it is determined that the current period is a pseudo-stationary period of LSTM model memory failure, and the image acquisition function of the UAV is locked.

[0043] It's important to note that an effective imaging window refers to the period during which the drone remains relatively stationary during flight for a duration exceeding the camera shutter speed. This window is characterized by very small fluctuations in the drone's attitude angular velocity during this time, ensuring clear and reliable images. This stability must be genuine, not an artifact created by model prediction errors.

[0044] In this context, the "pseudo-stationary window" is a key interfering factor. It refers to the LSTM model incorrectly generating a seemingly stable prediction curve when a strong disturbance is about to occur or the model's prediction fails. Although numerically this prediction curve meets the acquisition conditions, the drone may actually be experiencing severe shaking. This pseudo-window poses a significant risk in practical operation; if image acquisition is performed under these conditions, the captured images will be blurry.

[0045] Preferably, in one embodiment of the present invention, the specific method for identifying the continuous time period in which attitude fluctuations satisfy a preset static condition as the candidate prediction window length is as follows: Preset a fluctuation safety threshold In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. It depends on the specific implementation situation; For any adjacent prediction time in the predicted flight attitude data sequence of the UAV within a future time period, the absolute value of the difference between the predicted angular velocity data of the UAV in the K-axis direction at any adjacent prediction time is recorded as the change in angular velocity of the UAV in the K-axis direction at any adjacent prediction time; the summation and normalization of the changes in angular velocity of the UAV in the three axes at any adjacent prediction time is recorded as the change in attitude angular velocity at any adjacent prediction time. Calculate the change in attitude angular velocity between adjacent prediction times in the predicted flight attitude data sequence; ensure that the change in attitude angular velocity remains continuously below a preset fluctuation safety threshold. The time period is identified as a candidate prediction window with a relatively static posture, and the time span of this period is recorded as the length of the candidate prediction window.

[0046] Preferably, in one embodiment of the present invention, the specific method for dynamically correcting and compressing the candidate prediction window length using the prediction window confidence decay coefficient to obtain the effective prediction window length is as follows: The final effective prediction window length is the rounded-up product of the candidate prediction window length and the prediction window confidence decay coefficient of the LSTM model at the current time. The specific formula is as follows: In the formula, Indicates the effective prediction window length; Indicates the length of the candidate prediction window; This represents the confidence decay coefficient of the prediction window of the LSTM model at the current time. This indicates taking the value upwards.

[0047] Preferably, in one embodiment of the present invention, if the effective prediction window length is greater than or equal to the time required for a minimum exposure of the camera carried by the drone, the drone is controlled to perform image acquisition; if the effective prediction window length is less than the time required for a minimum exposure of the camera carried by the drone, the drone's image acquisition function is locked.

[0048] It should be noted that as the confidence decay coefficient of the prediction window decreases, it indicates that the UAV is more susceptible to environmental disturbances or that the LSTM model prediction has failed, resulting in a smaller effective prediction window length. When the effective prediction window length is shortened to less than the time required for a single minimum exposure by the camera, the UAV's image acquisition function is locked, forcing the system to abandon acquisition at highly unstable times. This process effectively solves the problems of "prediction lag" and "false triggering" in highly dynamic environments. Ensuring that each image acquisition occurs within a relatively stable environmental window improves image quality and reliability, ensuring higher credibility for the UAV during mission execution.

[0049] Please see Figure 3 The diagram illustrates a comparison of data acquisition execution strategies in an intelligent control system for UAV flight data acquisition. Within the volatile 400ms-600ms range, existing technologies still triggered two image captures, resulting in unusable images. However, the present invention, intercepted by confidence logic, intelligently skips this dangerous period, acquiring data only during stable periods before and after. This perfectly demonstrates that the present invention can effectively avoid motion blur and improve the quality of UAV flight data acquisition.

[0050] This concludes the embodiment.

[0051] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the patent of the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An intelligent control system for acquiring flight data from unmanned aerial vehicles (UAVs), characterized in that, The system includes: The data acquisition module is used to acquire the historical flight attitude data sequence of the UAV, as well as the time required for a single exposure of the camera carried by the UAV. The offset stability calculation module is used to extract angular velocity data from the historical flight attitude data sequence, calculate the degree of sudden change in the force on the UAV fuselage and the discrete volatility of the flight status data at the current moment; and combine the degree of sudden change in the force with the discrete volatility of the flight status data to obtain the deviation stability of the UAV's flight status at the current moment. The confidence acquisition module is used to input the historical flight attitude data sequence into the LSTM model to obtain the predicted flight attitude data sequence of the UAV in the future time period; based on the difference between the predicted flight attitude data and the actual flight attitude data of the UAV at the current moment, combined with the flight state deviation stability, to obtain the attitude prediction failure degree of the UAV at the current moment; and based on the attitude prediction failure degree, to obtain the prediction window confidence decay coefficient of the LSTM model at the current moment. The acquisition and control module is used to identify continuous time periods in the predicted flight attitude data sequence where attitude fluctuations meet preset static conditions as candidate prediction window lengths; dynamically correct and compress the candidate prediction window lengths using the prediction window confidence decay coefficient to obtain an effective prediction window length; if the effective prediction window length is greater than or equal to the time required for a minimum exposure of the UAV carrying the camera, the UAV is controlled to perform image acquisition within the corresponding effective prediction window; if the effective prediction window length is less than the time required for a minimum exposure of the UAV carrying the camera, it is determined that the current period is a pseudo-stationary period of LSTM model memory failure, and the image acquisition function of the UAV is locked.

2. The intelligent control system for UAV flight data acquisition according to claim 1, characterized in that, The calculation of the degree of sudden change in the stress on the UAV fuselage and the discrete fluctuation of the flight status data at the current moment includes: Preset a time parameter The closest time before the current moment A time window consisting of several moments is denoted as the backtracking time window of the current moment; Obtain the drone's position at all times within the backtracking time window. The standard deviation of the angular velocity data along the axis is denoted as the value of the UAV at the current moment. Discrete volatility of flight state data along the axial direction; The drone at the current moment The angular velocity data along the axis is compared with the previous moment when the UAV was The difference between the angular velocity data along the axis is denoted as the value of the angular velocity of the UAV at the current moment. The angular velocity difference along the axis; the absolute value of the difference between the current angular velocity difference and the previous angular velocity difference is denoted as the value of the difference between the UAV body and the angular velocity difference along the axis at the current moment. The degree of abrupt change in force along the axial direction.

3. The intelligent control system for UAV flight data acquisition according to claim 2, characterized in that, The step of combining the severity of the sudden force change with the discrete volatility of the flight state data to obtain the deviation of the UAV's flight state from stability at the current moment includes: The drone at the current moment Discrete fluctuations in flight state data along the axial direction and the current position of the UAV fuselage in... The product of the abrupt changes in force along the axial direction is denoted as the stress on the UAV fuselage at the current moment. The flight state in the axial direction deviates from the stability factor; The flight state deviation from stability of the UAV at the current moment is obtained by summing the deviation factors of the UAV's flight state in the three axes at the current moment.

4. The intelligent control system for UAV flight data acquisition according to claim 2, characterized in that, The process of obtaining the attitude prediction failure level of the UAV at the current moment includes: The predicted flight attitude data sequence includes predicted angular velocity data and predicted velocity data of the UAV in three axes. Based on the difference between the predicted flight attitude data and the actual flight attitude data of the UAV at the current moment, obtain the attitude difference value of the UAV at the current moment. Based on the attitude difference value at each moment in the backtracking time window, obtain the attitude prediction failure factor at each moment; By combining the attitude prediction failure factors at all moments in the backtracking time window with the flight state deviation stability, the degree of attitude prediction failure of the UAV at the current moment can be obtained.

5. The intelligent control system for UAV flight data acquisition according to claim 4, characterized in that, The process of obtaining the attitude difference value of the UAV at the current moment includes: The drone at the current moment The absolute value of the difference between the actual angular velocity data and the predicted angular velocity data in the axial direction is denoted as the angular velocity difference. The drone at the current moment The absolute value of the difference between the actual velocity data and the predicted velocity data along the axis is denoted as the velocity difference. The product of the velocity difference and the angular velocity difference is denoted as the value of the UAV's velocity at the current moment. Attitude difference factor along the axis; The summation and normalization of the attitude difference factors in the three axes of the UAV at the current moment are recorded as the attitude difference value of the UAV at the current moment.

6. The intelligent control system for UAV flight data acquisition according to claim 4, characterized in that, The acquisition of attitude prediction failure factors at each time step includes: The number of all moments in the backtracking time window is compared with the number of moments in the backtracking time window. The inverse proportional value of the difference between the index values ​​corresponding to the nth time point is denoted as the nth time point. Time weight of each moment; The first The time weight of the i-th moment and the i-th moment The product of the attitude difference values ​​of the UAV at time i is denoted as the i-th time. Attitude prediction failure factor at each time step.

7. The intelligent control system for UAV flight data acquisition according to claim 4, characterized in that, The step of combining the attitude prediction failure factors at all times within the backtracking time window with the flight state deviation stability to obtain the attitude prediction failure level of the UAV at the current time includes: The normalized value of the product of the mean of the attitude prediction failure factors at all times in the backtracking time window at the current time and the deviation of the UAV's flight state from stability at the current time is taken as the attitude prediction failure degree of the UAV at the current time.

8. The intelligent control system for UAV flight data acquisition according to claim 1, characterized in that, The step of obtaining the prediction window confidence decay coefficient of the LSTM model at the current moment based on the attitude prediction failure degree includes: Preset a model reliability threshold and a model failure threshold If the attitude prediction failure level of the UAV at the current moment is less than or equal to the model reliability threshold. Then This serves as the confidence decay coefficient for the prediction window of the LSTM model at the current moment; if the attitude prediction failure level of the UAV at the current moment is greater than the model reliability threshold... And less than the model failure threshold The attitude prediction failure level is directly used as the confidence decay coefficient of the prediction window of the LSTM model at the current moment; if the attitude prediction failure level of the UAV at the current moment is greater than or equal to the model failure threshold... Then This serves as the confidence decay coefficient for the prediction window of the LSTM model at the current moment.

9. The intelligent control system for UAV flight data acquisition according to claim 1, characterized in that, The step of dynamically correcting and compressing the candidate prediction window length using the prediction window confidence decay coefficient to obtain the effective prediction window length includes: The final effective prediction window length is the floor value of the product between the candidate prediction window length and the prediction window confidence decay coefficient of the LSTM model at the current time.

10. The intelligent control system for UAV flight data acquisition according to claim 1, characterized in that, The continuous time period during which attitude fluctuations satisfy a preset static condition is identified as the candidate prediction window length includes: Calculate the change in attitude angular velocity between adjacent prediction times in the predicted flight attitude data sequence; identify the time period in which the change in attitude angular velocity is continuously lower than a preset fluctuation safety threshold as a candidate prediction window where the attitude is relatively still, and record the time span of this period as the length of the candidate prediction window.

Images