A bird-detecting radar target tracking system and method based on deep learning

Abstract

The present invention relates to the technical field of bird-detecting radar target tracking, and specifically to a bird-detecting radar target tracking system and method based on deep learning. In the present invention, a data collection module acquires real-time and historical radar data and performs feature extraction to extract speed and acceleration data; a time window analysis module calculates a difference using the mean of historical speed and acceleration, and determines an initial prediction window length through autocorrelation analysis to solve the problem that traditional systems cannot adapt to dynamic changes in targets; a dynamic prediction window adjustment module dynamically adjusts the window length according to the real-time speed and acceleration difference, and the obtained real-time prediction window length enables the system to respond to changes more flexibly and improves the accuracy of prediction; a loss function optimization module calculates a weighting factor according to the real-time difference, and performs weighted processing on the loss function to improve the recognition of key change moments and enhance the overall prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of bird-detecting radar target tracking, and in particular to a bird-detecting radar target tracking system and method based on deep learning. Background Art

[0002] Traditional deep learning-based bird-detecting radar target tracking systems and methods usually face multiple technical problems.

[0003] First, there is a lack of effective evaluation of the periodicity and stability of the target's motion trajectory. Conventional methods may only rely on a fixed window length for prediction and cannot flexibly adapt to the dynamic characteristics of the target, resulting in a decrease in prediction accuracy when the motion trajectory changes significantly.

[0004] In addition, these systems often ignore changes in the real-time environment during the calculation process, fail to update or adjust the prediction window length in a timely manner, and cannot adapt to changes in the acceleration and speed of the target, resulting in the prediction results lagging behind the actual situation. In the setting of the loss function, traditional methods often fail to fully consider the significance of motion changes at different time points, and do not focus on time points with large changes, resulting in insufficient recognition and prediction capabilities of the model for key moments, affecting the overall prediction accuracy.

[0005] In the process of model training, failure to dynamically select appropriate historical data for iterative optimization may cause the model to rely on outdated data styles for a long time and be unable to cope with complex and changing environmental conditions. This traditional prediction system lacks a flexible adjustment mechanism, and its accuracy and effectiveness are restricted in the complex and changing background of bird migration, and cannot meet the needs of high-frequency dynamic updates.

[0006] Therefore, there is an urgent need for a forecasting strategy that can adapt to environmental changes and is more flexible and accurate to make up for the shortcomings of traditional systems. Summary of the invention

[0007] The purpose of the present invention is to provide a bird detection radar target tracking system and method based on deep learning to solve the problems raised in the above background technology.

[0008] To achieve the above object, one of the objects of the present invention is to provide a bird detection radar target tracking system based on deep learning, comprising a data collection module, a time window analysis module, a dynamic prediction window adjustment module and a loss function optimization module, wherein:

[0009] The data collection module acquires real-time radar data and historical radar data, and performs feature extraction on the real-time radar data and historical radar data, wherein the extracted results include the real-time speed and real-time acceleration at the current time point, and the historical speed and historical acceleration at each past time point;

[0010] The time window analysis module calculates the mean of the speed and acceleration in the historical radar data according to the historical speed and historical acceleration, and calculates the speed difference and acceleration difference of each past time point in the historical radar data according to the mean; and performs autocorrelation analysis according to the difference to determine the initial prediction window length;

[0011] The dynamic prediction window adjustment module obtains the speed difference and acceleration difference at the current time point in the real-time radar data according to the real-time speed and real-time acceleration at the current time point and the average of the speed and acceleration in the historical radar data; dynamically adjusts the initial prediction window length according to the difference to obtain the real-time prediction window length;

[0012] The loss function optimization module calculates the weighting factor at the current time point based on the speed difference and acceleration difference at the current time point in the real-time radar data, and performs weighted processing on the loss function in the deep learning algorithm model based on the weighting factor to optimize the loss function. The deep learning algorithm model is updated and predicted based on the real-time prediction window length and the optimized loss function.

[0013] As a further improvement of the technical solution, the process of extracting features of the data collection module specifically includes:

[0014] The radar sensor captures the target's motion information in real time. The collected real-time radar data includes the target's position coordinates at the current time point. ,in Indicates the current time point location;

[0015] Continuously record the target's motion trajectory and store the radar data at each past time point as historical radar data. The historical radar data includes the target position at the historical time point. ,in At every past time point location;

[0016] Real-time speed and real-time acceleration It is calculated based on the current and previous positions, and the formula is as follows:

[0017] , ,in is the time point interval;

[0018] Historical speed and historical acceleration It is calculated based on the position of each past time point and the position of the previous time point. The formula is as follows:

[0019] , ,in is the time point interval.

[0020] As a further improvement of the technical solution, the process of the time window analysis module calculating the velocity difference and acceleration difference at each past time point in the historical radar data specifically includes:

[0021] According to the historical speed and acceleration , calculate the mean speed in historical radar data and the mean value of acceleration ,in , , Indicates A past time point, Indicates the number of all past time points;

[0022] According to the calculated mean, the speed and acceleration difference of each past time point is obtained. For each past time point , speed difference and acceleration difference Calculation of , .

[0023] As a further improvement of the technical solution, the process of determining the initial prediction window length by the time window analysis module specifically includes:

[0024] Using the calculated speed difference and acceleration difference , use the autocorrelation function to perform autocorrelation analysis to determine the initial prediction window length , where the autocorrelation function is used to measure the correlation between different time points in the sequence. The calculation formula of the autocorrelation function is as follows:

[0025] ,in is the autocorrelation analysis of speed, is the speed difference The mean of is the lagged time point, representing the difference between the two time points compared in the autocorrelation function;

[0026] ,in is the autocorrelation analysis of acceleration, is the acceleration difference The mean of

[0027] According to the autocorrelation function , find the position where the autocorrelation function decays to near zero, and the lag time point at this position is the initial prediction window length. Taking into account the autocorrelation of velocity and acceleration, the minimum lag time point between them is taken as the initial prediction window length. .

[0028] As a further improvement of the technical solution, the process of determining the real-time prediction window length by the dynamic prediction window adjustment module specifically includes:

[0029] According to the real-time speed at the current time point and real-time acceleration , and the mean velocity in historical radar data and the mean value of acceleration , according to the formula as well as , calculate the speed difference at the current time point in the real-time radar data and acceleration difference ;

[0030] Set the adjustment threshold and adjustment coefficient, where the adjustment threshold includes the speed difference threshold and acceleration difference threshold , the adjustment factor includes the speed adjustment factor and acceleration adjustment factor ;

[0031] The initial prediction window length is adjusted using a linear scaling formula Adjust the formula as follows:

[0032] ,in is the adjusted initial prediction window length, which is used as the real-time prediction window length; ,in , .

[0033] As a further improvement of the technical solution, the process of optimizing the loss function by the loss function optimization module specifically includes:

[0034] According to the speed difference at the current time point in the real-time radar data and acceleration difference , according to the formula , calculate the weighting factor at the current time point ,in , , and is the normalization factor; It is an adjustment parameter used to control the influence of the weighting factor;

[0035] According to the weighting factor at the current time point , weighted processing is performed on the loss function in the deep learning algorithm model. In the loss function calculation formula, the error at the current time point is added to the weighted factor corresponding to the current time point. Weighted processing;

[0036] The weighted loss function is used to train the deep learning algorithm model, and the data corresponding to the historical radar data is selected as the training data for model update according to the real-time prediction window length. According to the updated deep learning algorithm model, the speed and acceleration of the radar data at future time points are predicted.

[0037] The second object of the present invention is to provide a method for using a bird detection radar target tracking system based on deep learning, comprising the following method steps:

[0038] S1. Acquire real-time radar data and historical radar data, and perform feature extraction on the real-time radar data and historical radar data, wherein the extracted results include the real-time speed and real-time acceleration at the current time point, and the historical speed and historical acceleration at each past time point;

[0039] S2. Calculate the mean of the speed and acceleration in the historical radar data according to the historical speed and historical acceleration, and calculate the speed difference and acceleration difference of each past time point in the historical radar data according to the mean; perform autocorrelation analysis according to the difference to determine the length of the initial prediction window;

[0040] S3, according to the real-time speed and real-time acceleration at the current time point, and the average of the speed and acceleration in the historical radar data, obtain the speed difference and acceleration difference at the current time point in the real-time radar data; according to the difference, dynamically adjust the initial prediction window length to obtain the real-time prediction window length;

[0041] S4. Calculate the weighting factor at the current time point based on the speed difference and acceleration difference at the current time point in the real-time radar data. Perform weighted processing on the loss function in the deep learning algorithm model based on the weighting factor to optimize the loss function. Update and predict the deep learning algorithm model based on the real-time prediction window length and the optimized loss function.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] A bird-detecting radar target tracking system and method based on deep learning, in the process of determining the initial prediction window length, the velocity difference and acceleration difference at each time point in the historical radar data are calculated through autocorrelation analysis, and these differences are used as the input of the autocorrelation function to evaluate the periodicity and stability of the historical data and determine the correlation between different time points; in the autocorrelation analysis, the position where the autocorrelation function decays to near zero is found, and the lag time point corresponding to this position is the initial prediction window length; through this method, the system effectively captures the historical style and change law of the target movement, and provides a benchmark for subsequent predictions, which ensures that the initial prediction window length can reflect the optimal time period of the target dynamic change, making the prediction model more targeted and effective;

[0044] In addition, the dynamic adjustment of the initial prediction window length involves the real-time intervention of the dynamic prediction window adjustment module, which uses the difference between the real-time speed and acceleration at the current time point and the average of the historical speed and acceleration; by setting the adjustment threshold and adjustment coefficient, it is determined whether the window length needs to be adjusted according to the difference between the speed and acceleration, and the linear scaling formula is used to achieve this adjustment; the application of this multiple real-time difference ensures that the system can flexibly respond to the instant changes in the target motion trajectory, so that the prediction window can be adaptively shrunk or expanded; as a result, the prediction stability and accuracy are improved in a dynamic environment, and can more accurately reflect the real motion trend;

[0045] In addition, the loss function optimization module uses the speed and acceleration differences at the current time point to calculate the weighting factor, and weights the error in the loss function of the deep learning model. The core of this strategy is to use the difference information to enhance the model's attention to time points with large changes, and to amplify the contribution of these points in the loss function by adjusting the normalization factor and weighting parameters. This process improves the model's ability to recognize and respond to key moments of change, so that it can optimize data with obvious changes during model training and updating, improve the prediction accuracy of the overall model, and combine with the real-time prediction window length to select appropriate historical radar data for model training, ensuring that the model is continuously optimized and accurately predicts the speed and acceleration of future time points. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0047] Figure 2 It is a schematic diagram of the overall module of the present invention.

[0048] In the figure: 100, data collection module; 200, time window analysis module; 300, dynamic prediction window adjustment module; 400, loss function optimization module. DETAILED DESCRIPTION

[0049] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Next, see Figure 1-Figure 2 One of the purposes of this embodiment is to provide a bird detection radar target tracking system based on deep learning, including a data collection module 100, a time window analysis module 200, a dynamic prediction window adjustment module 300 and a loss function optimization module 400.

[0051] The data collection module 100 obtains real-time radar data and historical radar data, and performs feature extraction on the real-time radar data and historical radar data. The extracted results include the real-time speed and real-time acceleration at the current time point, and the historical speed and historical acceleration at each past time point, specifically including:

[0052] The radar sensor captures the target's motion information in real time. The collected real-time radar data includes the target's position coordinates at the current time point. ,in Indicates the current time point location;

[0053] Continuously record the target's motion trajectory and store the radar data at each past time point as historical radar data. The historical radar data includes the target position at the historical time point. ,in At every past time point location;

[0054] Real-time speed and real-time acceleration It is calculated based on the current and previous positions, and the formula is as follows:

[0055] , ,in is the time point interval;

[0056] Historical speed and historical acceleration It is calculated based on the position of each past time point and the position of the previous time point. The formula is as follows:

[0057] , ,in is the time point interval.

[0058] The time window analysis module 200 calculates the mean of the speed and acceleration in the historical radar data according to the historical speed and historical acceleration, and calculates the speed difference and acceleration difference of each past time point in the historical radar data according to the mean; and performs autocorrelation analysis according to the difference to determine the initial prediction window length, which specifically includes:

[0059] According to the historical speed and acceleration , calculate the mean speed in historical radar data and the mean value of acceleration ,in , , Indicates A past time point, Indicates the number of all past time points;

[0060] According to the calculated mean, the speed and acceleration difference of each past time point is obtained. For each past time point , speed difference and acceleration difference Calculation of , ;

[0061] Using the calculated speed difference and acceleration difference , use the autocorrelation function to perform autocorrelation analysis to determine the initial prediction window length , where the autocorrelation function is used to measure the correlation between different time points in the sequence. The calculation formula of the autocorrelation function is as follows:

[0062] ,in is the autocorrelation analysis of speed, is the speed difference The mean of is the lagged time point, representing the difference between the two time points compared in the autocorrelation function;

[0063] ,in is the autocorrelation analysis of acceleration, is the acceleration difference The mean of

[0064] According to the autocorrelation function , find the position where the autocorrelation function decays to near zero, and the lag time point at this position is the initial prediction window length. Taking into account the autocorrelation of velocity and acceleration, the minimum lag time point between them is taken as the initial prediction window length. .

[0065] The dynamic prediction window adjustment module 300 obtains the speed difference and acceleration difference at the current time point in the real-time radar data according to the real-time speed and real-time acceleration at the current time point and the average of the speed and acceleration in the historical radar data; and dynamically adjusts the initial prediction window length according to the difference to obtain the real-time prediction window length, specifically including:

[0066] According to the real-time speed at the current time point and real-time acceleration , and the mean velocity in historical radar data and the mean value of acceleration , according to the formula as well as , calculate the speed difference at the current time point in the real-time radar data and acceleration difference ;

[0067] Set the adjustment threshold and adjustment coefficient, where the adjustment threshold includes the speed difference threshold and acceleration difference threshold , the adjustment factor includes the speed adjustment factor and acceleration adjustment factor ;

[0068] The initial prediction window length is adjusted using a linear scaling formula Adjust the formula as follows:

[0069] ,in is the adjusted initial prediction window length, which is used as the real-time prediction window length; ,in , .

[0070] The loss function optimization module 400 calculates the weighting factor at the current time point according to the speed difference and acceleration difference at the current time point in the real-time radar data, and performs weighted processing on the loss function in the deep learning algorithm model according to the weighting factor to optimize the loss function. The deep learning algorithm model is updated and predicted according to the real-time prediction window length and the optimized loss function, which specifically includes:

[0071] According to the speed difference at the current time point in the real-time radar data and acceleration difference , according to the formula , calculate the weighting factor at the current time point ,in , , and is the normalization factor; It is an adjustment parameter used to control the influence of the weighting factor;

[0072] According to the weighting factor at the current time point , weighted processing is performed on the loss function in the deep learning algorithm model. In the loss function calculation formula, the error at the current time point is added to the weighted factor corresponding to the current time point. Weighted processing, through this weighting, a weighted loss function is obtained, so that when the change is more significant, the corresponding error contributes more to the total loss;

[0073] The weighted loss function is used to train the deep learning algorithm model, and the data corresponding to the historical radar data is selected as the training data for model update according to the real-time prediction window length. According to the updated deep learning algorithm model, the speed and acceleration of the radar data at future time points are predicted.

[0074] The second purpose of this embodiment is to provide a method for using a bird detection radar target tracking system based on deep learning, including the following method steps:

[0075] S1. Acquire real-time radar data and historical radar data, and perform feature extraction on the real-time radar data and historical radar data, wherein the extracted results include the real-time speed and real-time acceleration at the current time point, and the historical speed and historical acceleration at each past time point;

[0076] S2. Calculate the mean of the speed and acceleration in the historical radar data according to the historical speed and historical acceleration, and calculate the speed difference and acceleration difference of each past time point in the historical radar data according to the mean; perform autocorrelation analysis according to the difference to determine the length of the initial prediction window;

[0077] S3, according to the real-time speed and real-time acceleration at the current time point, and the average of the speed and acceleration in the historical radar data, obtain the speed difference and acceleration difference at the current time point in the real-time radar data; according to the difference, dynamically adjust the initial prediction window length to obtain the real-time prediction window length;

[0078] S4. Calculate the weighting factor at the current time point based on the speed difference and acceleration difference at the current time point in the real-time radar data. Perform weighted processing on the loss function in the deep learning algorithm model based on the weighting factor to optimize the loss function. Update and predict the deep learning algorithm model based on the real-time prediction window length and the optimized loss function.

[0079] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A bird detection radar target tracking system based on deep learning, characterized in that: The method comprises a data collection module (100), a time window analysis module (200), a dynamic prediction window adjustment module (300) and a loss function optimization module (400), wherein: The data collection module (100) acquires real-time radar data and historical radar data, and performs feature extraction on the real-time radar data and the historical radar data, wherein the extraction results include the real-time speed and real-time acceleration at the current time point, and the historical speed and historical acceleration at each past time point; The time window analysis module (200) calculates the mean of the speed and acceleration in the historical radar data based on the historical speed and historical acceleration, and calculates the speed difference and acceleration difference at each past time point in the historical radar data based on the mean; and performs autocorrelation analysis based on the difference to determine the initial prediction window length; The dynamic prediction window adjustment module (300) obtains the speed difference and acceleration difference at the current time point in the real-time radar data according to the real-time speed and real-time acceleration at the current time point and the average of the speed and acceleration in the historical radar data; dynamically adjusts the initial prediction window length according to the difference to obtain the real-time prediction window length; The loss function optimization module (400) calculates a weighting factor at the current time point based on the speed difference and acceleration difference at the current time point in the real-time radar data, and performs weighted processing on the loss function in the deep learning algorithm model based on the weighting factor to optimize the loss function, and updates and predicts the deep learning algorithm model based on the real-time prediction window length and the optimized loss function.

2. The bird detection radar target tracking system based on deep learning according to claim 1 is characterized in that: The feature extraction process of the data collection module (100) specifically includes: The radar sensor captures the target's motion information in real time. The collected real-time radar data includes the target's position coordinates at the current time point. ,in Indicates the current time point location; Continuously record the target's motion trajectory and store the radar data at each past time point as historical radar data. The historical radar data includes the target position at the historical time point. ,in At every past time point location; Real-time speed and real-time acceleration It is calculated based on the current and previous positions, and the formula is as follows: , ,in is the time point interval; Historical speed and historical acceleration It is calculated based on the position of each past time point and the position of the previous time point. The formula is as follows: , ,in is the time point interval.

3. The bird detection radar target tracking system based on deep learning according to claim 2 is characterized in that: The process of the time window analysis module (200) calculating the velocity difference and acceleration difference at each past time point in the historical radar data specifically includes: According to the historical speed and acceleration , calculate the mean speed in historical radar data and the mean value of acceleration ,in , , Indicates A past time point, Indicates the number of all past time points; According to the calculated mean, the speed and acceleration difference of each past time point is obtained. For each past time point , speed difference and acceleration difference Calculation of , .

4. The bird detection radar target tracking system based on deep learning according to claim 3 is characterized in that: The process of determining the initial prediction window length by the time window analysis module (200) specifically includes: Using the calculated speed difference and acceleration difference , use the autocorrelation function to perform autocorrelation analysis to determine the initial prediction window length , where the autocorrelation function is used to measure the correlation between different time points in the sequence. The calculation formula of the autocorrelation function is as follows: ,in is the autocorrelation analysis of speed, is the speed difference The mean of is the lagged time point, representing the difference between the two time points compared in the autocorrelation function; ,in is the autocorrelation analysis of acceleration, is the acceleration difference The mean of According to the autocorrelation function , find the position where the autocorrelation function decays to near zero, and the lag time point at this position is the initial prediction window length. Taking into account the autocorrelation of velocity and acceleration, the minimum lag time point between them is taken as the initial prediction window length. .

5. The bird detection radar target tracking system based on deep learning according to claim 4 is characterized in that: The process of determining the real-time prediction window length by the dynamic prediction window adjustment module (300) specifically includes: According to the real-time speed at the current time point and real-time acceleration , and the mean velocity in historical radar data and the mean value of acceleration , according to the formula as well as , calculate the speed difference at the current time point in the real-time radar data and acceleration difference ; Set the adjustment threshold and adjustment coefficient, where the adjustment threshold includes the speed difference threshold and acceleration difference threshold , the adjustment factor includes the speed adjustment factor and acceleration adjustment factor ; The initial prediction window length is adjusted using a linear scaling formula Adjust the formula as follows: ,in is the adjusted initial prediction window length, which is used as the real-time prediction window length; ,in , .

6. The bird detection radar target tracking system based on deep learning according to claim 5 is characterized in that: The process of optimizing the loss function by the loss function optimization module (400) specifically includes: According to the speed difference at the current time point in the real-time radar data and acceleration difference , according to the formula , calculate the weighting factor at the current time point ,in , , and is the normalization factor; It is an adjustment parameter used to control the influence of the weighting factor; According to the weighting factor at the current time point , weighted processing is performed on the loss function in the deep learning algorithm model. In the loss function calculation formula, the error at the current time point is added to the weighted factor corresponding to the current time point. Weighted processing; The weighted loss function is used to train the deep learning algorithm model, and the data corresponding to the historical radar data is selected as the training data for model update according to the real-time prediction window length. According to the updated deep learning algorithm model, the speed and acceleration of the radar data at future time points are predicted.

7. A method for using the bird detection radar target tracking system based on deep learning according to any one of claims 1 to 6, characterized in that: The method comprises the following steps: S1. Acquire real-time radar data and historical radar data, and perform feature extraction on the real-time radar data and historical radar data, wherein the extracted results include the real-time speed and real-time acceleration at the current time point, and the historical speed and historical acceleration at each past time point; S2. Calculate the mean of the speed and acceleration in the historical radar data according to the historical speed and historical acceleration, and calculate the speed difference and acceleration difference of each past time point in the historical radar data according to the mean; perform autocorrelation analysis according to the difference to determine the length of the initial prediction window; S3, according to the real-time speed and real-time acceleration at the current time point, and the average of the speed and acceleration in the historical radar data, obtain the speed difference and acceleration difference at the current time point in the real-time radar data; according to the difference, dynamically adjust the initial prediction window length to obtain the real-time prediction window length; S4. Calculate the weighting factor at the current time point based on the speed difference and acceleration difference at the current time point in the real-time radar data. Perform weighted processing on the loss function in the deep learning algorithm model based on the weighting factor to optimize the loss function. Update and predict the deep learning algorithm model based on the real-time prediction window length and the optimized loss function.

Images