Navigation resolving method and device for collaborative utilization of multi-source sensor data

Abstract

The invention discloses a navigation resolving method and device for collaborative utilization of multi-source sensor data. The method comprises the following steps: firstly, carrying out state definition and initialization; extracting features based on multi-source sensor data and constructing feature vectors; clustering the feature vectors by using a clustering algorithm to identify different underwater scene categories; for each scene category, establishing a sensor data weight distribution function by using the trained machine learning model; and finally, dynamically distributing the weight of each sensor according to the current scene, and completing navigation calculation by adopting a weighted fusion algorithm. By introducing a dynamic weight fusion mechanism based on scene classification, the limitation of a traditional fixed weight fusion strategy in different underwater environments is overcome, and the precision, environmental adaptability and overall reliability of a navigation system are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the technical field of underwater navigation, and particularly relates to a navigation calculation method and device that utilizes multi-source sensor data collaboratively. Background Technology

[0002] In the field of underwater navigation, existing technologies typically involve simply fusing data from various sensors. However, different underwater scenarios (shallow sea, deep sea, nearshore) have different characteristics, and simple data fusion methods cannot fully leverage the performance advantages of each sensor, thus limiting the accuracy and reliability of navigation solutions.

[0003] Different underwater environments, such as shallow seas, deep seas, nearshore areas, and caves, have different characteristics. These characteristics affect the performance of sensors such as DVL (Device Vibration Volume), side-scan sonar, and IMU (Infrared Detection Unit) to varying degrees. Specifically, DVL calculates the velocity of a vehicle relative to the seabed by measuring acoustic signals. In the deep-sea environment, due to the long propagation distance of acoustic waves, signal attenuation is relatively severe, and the measurement accuracy may be reduced due to complex ocean currents. In shallow nearshore areas, there may be many interference sources, such as ships and waves, which can also affect the accuracy of DVL measurements. Side-scan sonar is used to acquire seabed topographic information. In areas with complex topography, such as underwater mountains and trenches, it can provide rich texture information, which greatly helps navigation. However, in relatively flat areas such as deep-sea plains, the information acquired by side-scan sonar changes relatively little, and its contribution to navigation is limited. IMU mainly measures the vehicle's acceleration and angular velocity, and can provide relatively accurate attitude and motion information in a short period of time. However, cumulative errors can occur over prolonged use, and these errors can increase rapidly in scenarios with strong currents or violent carrier motion. Existing methods do not differentiate between different scenarios, employing a fixed-weight fusion strategy without adjusting the weights of DVL, side-scan sonar, and IMU based on the scenario. Therefore, there is room for further improvement in navigation and positioning accuracy. Consequently, a novel multi-source data collaborative utilization strategy is urgently needed to enhance the accuracy and reliability of underwater navigation. Summary of the Invention

[0004] In view of this, the present invention provides a navigation solution method and apparatus for collaborative utilization of multi-source sensor data, which can dynamically allocate the weight of each sensor data in the navigation solution according to the characteristics of different underwater scenes and the performance characteristics of sensors, thereby improving the accuracy and reliability of underwater navigation.

[0005] The technical solution for implementing the present invention is as follows: A navigation solution method utilizing multi-source sensor data collaboratively includes the following steps: Step 1, State Definition and Initialization: Define the state vector of the navigation system and initialize the state vector and covariance matrix; Step 2, Underwater Scene Recognition: Extract features based on multi-source sensor data, wherein the multi-source sensors include at least DVL, side-scan sonar, and IMU, and the features include at least the sonar image texture complexity calculated based on the gray-level co-occurrence matrix and the DVL intensity signal, and construct a feature vector; Step 3, Underwater Scene Clustering: Cluster the feature vector using a clustering algorithm to identify different underwater scene categories; Step 4, Weight Allocation Model Establishment: For each identified underwater scene category, establish a sensor data weight allocation function using a machine learning model trained on sample data; Step 5, Weighted Data Fusion: Dynamically allocate the weights of each sensor data according to the weight allocation function corresponding to the current scene category, and perform navigation solution using a weighted fusion algorithm.

[0006] Furthermore, the clustering algorithm is the K-means clustering algorithm.

[0007] Furthermore, the machine learning model is a support vector regression model.

[0008] Furthermore, the weighted fusion algorithm is an extended Kalman filter; in the update step of the extended Kalman filter, the noise covariance matrix is ​​measured. The sensors are weighted and combined according to their dynamic weights. The combination formula is as follows:

[0009] , and These are the measurement noise covariance matrices for DVL, sonar, and IMU, respectively. , and These are the corresponding dynamic weights.

[0010] Furthermore, in the weighted data fusion step, the measurement vector used for fusion is formed by weighting and combining the measurement vectors of each sensor according to their corresponding dynamic weights.

[0011] A navigation calculation device for collaborative utilization of multi-source sensor data includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0012] Beneficial effects: 1. This invention overcomes the performance limitations of traditional fixed-weight fusion strategies in different scenarios by dynamically adjusting the weights of data fusion from each sensor based on the real-time identified underwater scene, thereby achieving more accurate navigation solutions.

[0013] 2. This invention combines scene recognition with data fusion, enabling the navigation system to automatically adapt to the characteristics of different underwater environments such as shallow sea, deep sea, and nearshore, making full use of the performance advantages of each sensor in different scenarios, and improving the overall robustness and applicability of the system.

[0014] 3. By intelligently allocating sensor weights, the system can automatically reduce the contribution of unreliable data in scenarios where sensor signal quality deteriorates or environmental interference is severe, thus optimizing the utilization of limited sensor resources, effectively suppressing cumulative errors, and improving the reliability of long-term navigation. Attached Figure Description

[0015] Figure 1 This is a flowchart of the navigation solution method for collaborative utilization of multi-source sensor data according to the present invention. Detailed Implementation

[0016] This invention provides a navigation solution method and apparatus for collaborative utilization of multi-source sensor data. The specific process of the method is as follows: Figure 1 As shown, by Figure 1 As can be seen, the method of the present invention mainly includes five steps: state definition and initialization, underwater scene recognition, underwater scene clustering, weight allocation model establishment, and weighted data fusion, which are described in detail below.

[0017] 1. State definition and initialization Define state vector , representing position, velocity, attitude angle, and angular velocity, respectively. Initialize the state vector. Covariance Matrix .

[0018] 2. Underwater scene recognition For sonar, calculate its image texture complexity. This includes, but is not limited to, using the Gray-Level Co-occurrence Matrix (GLCM) method. Taking GLCM as an example, the gray-level co-occurrence matrix is ​​first calculated. For each pixel in the image Calculate its gray-level co-occurrence relationship with neighboring pixels at specific directions and distances. Then, based on... Calculating texture complexity Features, including but not limited to For DVL, the intensity signal output by the DVL device can be read directly. These features are combined into a feature vector. .

[0019] 3. Underwater Scene Clustering Clustering algorithms are used to analyze feature vectors. Clustering is performed. This includes, but is not limited to, using the K-means method. Taking the K-means method as an example, first, random initialization is performed. Cluster centers Then for each feature vector Calculate the Euclidean distance between it and the cluster center.

[0020] (1) in, and These are the feature vectors. and cluster center The Each component. Then... Assign it to the category of the nearest cluster center, i.e. (2) For feature vectors The category labels are then updated. The cluster centers are then updated for each category. New cluster center ,in It is a category The number of samples in the cluster. Repeat the above steps until the cluster centers no longer change or other stopping conditions are met.

[0021] 4. Establishment of the weight allocation model The weighting model can be constructed using a support vector regression (SVR) model. Specifically, for each cluster, the weighting model is determined by the following parameters: Collect feature vector samples belonging to this scene ( (This refers to the number of samples in the scene) and the corresponding true values ​​of the DVL data weights. (This can be obtained through expert annotation). Next, the model is trained using Support Vector Regression (SVR). The objective function of SVR is... (3) The constraints are

[0022] (4)

[0023] in, It is a weight vector. It is a bias term. and It is a slack variable. It is a penalty parameter. It is an insensitive loss parameter. It is a function that maps input characteristics to a high-dimensional space. By solving the above optimization problem, the model parameters are obtained. (5) Thus, the weight allocation function is obtained. (6) Similarly, the weights of sonar data and IMU data can be obtained as follows: and .

[0024] 5. Weighted Data Fusion Data fusion is performed using an extended Kalman filter (EKF) with dynamic weights.

[0025] 1) The system state equation is: (7) in, It is a state vector, including position. ,speed ,attitude And the zero bias of the IMU and ; It is the state transition matrix. It is a noise-driven matrix. It is system noise.

[0026] 2) The measurement equation is: (8) in, It is a measurement vector, used to measure velocity using DVL. Location of terrain feature points extracted by side-scan sonar composition; It is a measurement matrix. It measures noise.

[0027] 3) EKF prediction steps: (9) in, For prior state estimation, It is a priori estimate of covariance. It is the system noise covariance.

[0028] 3) EKF update steps: First, calculate the measurement noise covariance matrix. The weighted combination is performed based on the weights of each sensor: (10) , , These are the measurement noise covariance matrices for DVL, sonar, and IMU, respectively. Then, the Kalman gain is calculated. : (11) Next, update the state vector: (12) in, It is the fused measurement vector. , , , These are the measurement vectors for DVL, sonar, and IMU, respectively. Finally, the covariance matrix is ​​updated. : (13) Through the above steps, scene recognition and weight allocation are integrated with the EKF algorithm, enabling more accurate navigation and positioning solutions for different underwater scenarios.

[0029] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A navigation solution method that utilizes multi-source sensor data collaboratively, characterized in that, Includes the following steps: Step 1, State Definition and Initialization: Define the state vector of the navigation system and initialize the state vector and covariance matrix; Step 2, Underwater Scene Recognition: Extract features based on multi-source sensor data, where the multi-source sensors include at least DVL, side-scan sonar, and IMU. The features include at least the sonar image texture complexity calculated based on the gray-level co-occurrence matrix and the DVL intensity signal, and construct a feature vector; Step 3, Underwater Scene Clustering: Cluster the feature vector using a clustering algorithm to identify different underwater scene categories; Step 4, Weight Allocation Model Establishment: For each identified underwater scene category, establish a sensor data weight allocation function using a machine learning model trained on sample data; Step 5, Weighted Data Fusion: Dynamically allocate the weights of each sensor data according to the weight allocation function corresponding to the current scene category, and use a weighted fusion algorithm for navigation calculation.

2. The method according to claim 1, characterized in that, The clustering algorithm is the K-means clustering algorithm.

3. The method according to claim 1, characterized in that, The machine learning model is a support vector regression model.

4. The method according to any one of claims 1-3, characterized in that, The weighted fusion algorithm is an extended Kalman filter; in the update step of the extended Kalman filter, the noise covariance matrix is ​​measured. The sensors are weighted and combined according to their dynamic weights. The combination formula is as follows: , and These are the measurement noise covariance matrices for DVL, sonar, and IMU, respectively. , and These are the corresponding dynamic weights.

5. The method according to claim 4, characterized in that, In the weighted data fusion step, the measurement vector used for fusion is formed by weighting and combining the measurement vectors of each sensor according to their corresponding dynamic weights.

6. A navigation calculation device for collaborative utilization of multi-source sensor data, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the program to implement the method as described in any one of claims 1 to 5.

Images