A full coverage target search method for autonomous underwater vehicles

Abstract

The application discloses a full-coverage target search method for an autonomous underwater vehicle and belongs to the technical field of autonomous underwater vehicle control, and aims to solve the problems in the prior art. The application comprises the following steps: downloading an elevation map of a task sea area and inputting the elevation map into an AUV; selecting a grid scale according to a detection sensor operation radius and an AUV turning radius, performing grid processing on the elevation map, and obtaining a grid map; representing the grid map obtained in step two based on a GBNN algorithm, obtaining a 2D neural network, and performing initialization iteration on the 2D neural network to obtain an iterated 2D neural network; confirming the navigable areas affected by obstacles in the iterated 2D neural network, taking the part of the navigable areas as special areas, and reassigning values to neurons in the special areas; performing full-coverage path planning on the AUV; and based on the full-coverage path, optimizing the planned path in real time based on a turning optimization strategy and a GBNN-based escape dead point method to obtain a final navigation path of the AUV.

Description

Technical Field

[0001] This invention belongs to the field of autonomous underwater vehicle control technology, specifically relating to a full-coverage target search method for autonomous underwater vehicles. Background Technology

[0002] Autonomous underwater vehicles (AUVs) are widely used in marine geomorphological mapping, marine natural resource exploration, underwater archaeology, and underwater search and rescue due to their high maneuverability, flexibility, and efficiency. However, during detailed searches of specific sea areas, the low coverage and energy consumption of path planning remain challenges due to underwater obstacles and energy constraints. Furthermore, to ensure stable imaging by detection equipment, the AUV's operating depth should not be frequently changed. Therefore, constant-depth, time-coverage path planning with high safety, stability, and efficiency is a core issue in AUV applications.

[0003] Random search algorithms are among the earliest cover-type path planning algorithms. The agent moves randomly within the search domain. While relatively simple, these algorithms struggle to escape deadlocks in complex environments, resulting in low search efficiency. In recent years, heuristic methods have seen rapid development, with genetic algorithms and artificial neural networks widely used in path planning. However, heuristic algorithms suffer from high computational cost and slow convergence. Bio-inspired neural networks can also be used to solve cover-type path planning problems. Representative algorithms include Bio-Inspired Neural Network (BINN) and Glasius Bio-Inspired Neural Network (GBNN), but these algorithms also suffer from long path planning times. Cover-type path planning algorithms based on energy compensation and obstacle vectorization are based on template methods, but these algorithms are not suitable for marine robots with omnidirectional turning capabilities. Summary of the Invention

[0004] The purpose of this invention is to propose a comprehensive target search method for autonomous underwater vehicles, which solves the problems of existing technologies, such as difficulty in getting rid of deadlock points, low search efficiency, large computational load, slow convergence speed, long path planning time, and unsuitability for marine robots with omnidirectional turning capabilities.

[0005] To achieve the above objectives, the present invention provides a full-coverage target search method for autonomous underwater vehicles, comprising the following steps:

[0006] Step 1: Download the elevation map of the mission area and import it into the AUV;

[0007] Step 2: Select the grid scale based on the operating radius of the detection sensor and the turning radius of the AUV, and rasterize the elevation map obtained in Step 1 to obtain a raster map; the relationship between the selected grid scale L and the turning radius R1 of the AUV and the operating radius R2 of the detection sensor is as follows:

[0008]

[0009] Step 3: Represent the grid map obtained in Step 2 using the GBNN (Gaussian Biological Inspired Neural Network) algorithm to obtain a 2D neural network;

[0010] Step 4: Initialize and iterate the 2D neural network obtained in Step 3 to obtain the iterated 2D neural network;

[0011] Step 5: Identify the navigable area affected by obstacles in the iterated 2D neural network; this navigable area is designated as a special region.

[0012] Step Six: Reassign values ​​to neurons in the special regions identified in Step Five;

[0013] Step 7: Based on the iterative 2D neural network obtained in Step 4 and the special regions reassigned in Step 6, the AUV is controlled to navigate in a "bow" shaped path in the non-special regions of the neural network. When entering a special region, the AUV is guided by the distribution of neuron activity values ​​in that region, thus completing the full-coverage path planning for the AUV. Based on the obtained full-coverage path, the full-coverage path is optimized in real time using a steering optimization strategy and a GBNN-based deadlock escape method to obtain the final navigation path of the AUV.

[0014] The special region mentioned in step five includes the area not covered in the AUV reverse extension direction; specifically, the area not covered in the AUV reverse extension direction is: the area from the AUV to (x A ,y A When (x) A -1,y A ) and (x A +1,y A When the neuronal activity values ​​of (x) are all greater than or equal to 1, then (x) A -1,y A () is the seed point, in [1, x) A The navigable connected area formed within the region. The extension direction is the horizontal direction of movement of the AUV. In this invention, taking the extension direction of the AUV as an example (from left to right), the reverse extension direction is from right to left.

[0015] The special area mentioned in step five also includes the area containing concave obstacles in the traversal extension direction; the area containing concave obstacles in the traversal extension direction includes: the area adjacent to the AUV when it is flying straight, the area adjacent to the AUV when it turns at the obstacle boundary, and the area adjacent to the AUV when it turns at the operation boundary.

[0016] The specific area adjacent to the AUV during direct flight is:

[0017] When the AUV's heading is 90° and it is sailing to (x A ,y A When (x) A +1,y A activity value And (x) A +1,y A -1) activity value When, then respectively with (x A +1,y A ) and (x A +1,y A -1) Construct navigable connected regions and obstacle connected regions for seed points respectively;

[0018] When the AUV's heading is -90° and it is sailing to (x A ,y A When (x) A +1,y A Neuronal activity value at ) and When, then respectively with (x A +1,y A ) and (x A +1,y A +1) Construct navigable connected regions and obstacle connected regions for the seed point;

[0019] If the navigable connectivity is contained within the obstacle connectivity, the navigable connectivity is recorded as a special area as the adjacent area when the AUV is flying straight.

[0020] If a navigable connected region is not contained within an obstacle connected region, then the navigable connected region is not considered a special region.

[0021] The specific area adjacent to the obstacle boundary when the AUV turns is as follows: when the AUV is traveling at a heading of -90 and reaches the obstacle boundary, the AUV turns and travels to (x A ,y A ),like Then (x) A +1,y A Construct a navigable connected region with (x) as the seed point of the navigable region, and use (x) as the seed point of the navigable region. A +1,y A-1) Construct connected domains for obstacle regions based on seed points;

[0022] When the AUV is traveling at a heading of 90° and reaches the boundary of the obstacle, it turns and travels to (x A ,y A ),like Then (x) A +1,y A Construct a navigable connected region with (x) as the seed point of the navigable region, and use (x) as the seed point of the navigable region. A +1,y A +1) constructs the obstacle connected domain for the seed point of the obstacle region;

[0023] If the navigable connected region is contained within the obstacle connected region, the navigable connected region is recorded as a special region as the area immediately adjacent to the AUV when it turns at the obstacle boundary.

[0024] If a navigable connected region is not contained within an obstacle connected region, then the navigable connected region is not considered a special region.

[0025] The specific area adjacent to the AUV when turning at the operational boundary is as follows: when the AUV is traveling at a heading of 90° or -90° to the boundary of the detection mission, the AUV turns and travels to (x) A ,y A ) position, if Then (x) A +1,y A Construct a navigable connected region using the seed point of the navigable region as the y-th point. A Column distance (x) A +1,y A The nearest obstacle grid is used as the seed point to construct the obstacle connected domain; the obstacle boundary on the side of the adjacent navigable connected domain is extracted by the edge detection method;

[0026] If the semi-closed area formed by the extracted obstacle boundary and the mission sea area boundary contains a navigable connected area, then the navigable connected area is recorded as a special area as the area adjacent to the AUV when it turns at the operational boundary.

[0027] If the semi-closed area formed by the extracted obstacle boundary and the mission sea area boundary does not contain a navigable connected domain, then the navigable connected domain is not considered a special area.

[0028] The method for reassigning neuronal values ​​in the special region described in step six is ​​as follows:

[0029] First, define the gradient direction for increasing activity values; both the layer index and the neuron index in each layer increase along the gradient direction; the neuron activity value increases along the gradient direction, and the model for calculating the maximum activity value in each layer is as follows:

[0030] V(n) = 1 + η·ρ (n-1 / 2) n = 1, 2, ..., N

[0031] Where: V(n) represents the maximum activity value of neurons in the nth layer; N represents the maximum number of layers in the region; η and ρ are positive numbers;

[0032] The model for calculating the remaining neuron activity in each layer is as follows:

[0033]

[0034]

[0035]

[0036]

[0037] Where: V(i,n) represents the activity value of the i-th neuron in the n-th layer, M n This indicates the number of neurons in each layer.

[0038] The path planning calculation method in step seven is as follows: After iteration, neurons in the non-special regions of the 2D neural network guide the AUV to navigate in a "bow" shaped path. When the AUV reaches a special region, it is guided by the reassigned neuron activity value distribution. The AUV navigates from the neuron with the smallest activity value to the neuron with the largest activity value in the special region, completing the navigation in the special region. During the navigation, if the AUV enters a deadlock point neuron, a deadlock escape method based on GBNN is used to help the AUV escape the deadlock point. If multiple neurons are available for navigation when the AUV turns, a turning optimization strategy is used to optimize the AUV path.

[0039] The steering optimization strategy specifically includes the following steps:

[0040] Step 1: Using the current position of the AUV (x A ,y A Based on the distribution of neuron activity values ​​within an eight-neighbor area centered on , the next navigation position (x) of the AUV is calculated. P ,y P );

[0041] Step 2: Determine (x) P ,y P +1) and (x P ,y P -1) Are all neurons unsearched? If yes, proceed to step three; otherwise, the AUV navigates to position (x). P ,y P )

[0042] Step 3: Using (x)P ,y P +1) and (x P ,y P -1) is the seed point for the navigable region in the region {(x,y)|x A =x,y A +1≤y≤y max Constructing a connected component Ω within the} up and Ω down ;

[0043] Step 4: For Ω respectively up and Ω down The number of neurons within the turnup and turndown are counted as num_turnup and num_turndown, respectively.

[0044] Step 5: AUV prioritizes searching connected regions with fewer neurons. After the first search, the search is repeated once to search another connected region.

[0045] The GBNN-based method for escaping deadlock points specifically includes the following steps:

[0046] Step 1: Using the current position of the AUV (x A ,y A Using ) as the center, determine whether there are neurons with an activity value ≥1 in a neighborhood of radius n = 2. If yes, proceed to step three; otherwise, proceed to step two.

[0047] Step 2: Increment the value of n by one, and in (x A ,y A Using ) as the center, determine whether there are neurons with an activity value ≥1 in a neighborhood of radius n. If yes, proceed to step three; otherwise, repeat step two.

[0048] Step 3, in (x A ,y A Centered on the target, within a neighborhood of radius n, targeting unsearched neurons, the GBNN method is used to obtain an effective path from the current position of the AUV to the target position.

[0049] The beneficial effects of this invention are as follows: The comprehensive target search method for autonomous underwater vehicles (AUVs) is based on a Gaussian bio-inspired neural network model. By extracting navigable areas affected by obstacles as special regions and reassigning the activity values ​​of neurons in these regions, the method guides AUVs to achieve high-coverage navigation within the mission area. Simultaneously, during navigation, the AUV can escape deadlock points using the GBNN algorithm and select the optimal steering neuron through a steering optimization strategy. The AUV navigation path planned by this method has a lower path repetition rate, fewer deadlocks, and fewer turns, which is beneficial for AUVs to efficiently perform refined search tasks. Attached Figure Description

[0050] Figure 1 This is a flowchart of a full-coverage target search method for an autonomous underwater vehicle according to the present invention;

[0051] Figure 2 This invention relates to a grid map used in a full-coverage target search method for autonomous underwater vehicles.

[0052] Figure 3 This is a diagram illustrating the process of mapping a grid map to a 2D neural network in a comprehensive target search method for autonomous underwater vehicles according to the present invention.

[0053] Figure 4 This is a completed UARED constructed in the full-coverage target search method for autonomous underwater vehicles of the present invention;

[0054] Figure 5 This is a schematic diagram of neuron index values ​​in a full-coverage target search method for autonomous underwater vehicles according to the present invention;

[0055] Figure 6 This is a schematic diagram of the adjacent area when the AUV is traveling straight with a heading angle of -90° in a full-coverage target search method for autonomous underwater vehicles according to the present invention.

[0056] Figure 7 This is a schematic diagram of the adjacent area when the AUV is traveling straight at a heading angle of 90° in a full-coverage target search method for autonomous underwater vehicles according to the present invention.

[0057] Figure 8 This is a schematic diagram of the area adjacent to an obstacle boundary when the AUV turns at a heading angle of -90° in a full-coverage target search method for an autonomous underwater vehicle according to the present invention.

[0058] Figure 9This is a schematic diagram of the area adjacent to an obstacle boundary when the AUV turns at a 90° heading angle in a full-coverage target search method for an autonomous underwater vehicle according to the present invention.

[0059] Figure 10 This is a schematic diagram of the area adjacent to the AUV when the AUV turns at the operation boundary in a full-coverage target search method for autonomous underwater vehicles according to the present invention, where the AUV's heading angle is -90°.

[0060] Figure 11 This is a schematic diagram of the area adjacent to the AUV when the AUV turns at the operation boundary in a full-coverage target search method for autonomous underwater vehicles according to the present invention, where the AUV's heading angle is 90°.

[0061] Figure 12 This is a schematic diagram illustrating the multiple candidate steering neurons in a full-coverage target search method for autonomous underwater vehicles (AUVs) according to the present invention. Detailed Implementation

[0062] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0063] See Figures 1-12 The present invention provides a method for full-coverage target search for autonomous underwater vehicles, comprising the following steps:

[0064] Step 1: Download the elevation map of the mission area and import it into the AUV;

[0065] Step 2: Select the grid scale based on the operating radius of the detection sensor and the turning radius of the AUV, and rasterize the elevation map obtained in Step 1 to obtain a raster map; the relationship between the selected grid scale L and the turning radius R1 of the AUV and the operating radius R2 of the detection sensor is as follows:

[0066]

[0067] Step 3: Represent the grid map obtained in Step 2 using the GBNN (Gaussian Biological Inspired Neural Network) algorithm to obtain a 2D neural network;

[0068] Step 4: Initialize and iterate the 2D neural network obtained in Step 3 to obtain the iterated 2D neural network;

[0069] Step 5: Identify the navigable area affected by obstacles in the iterated 2D neural network; this navigable area is designated as a special region.

[0070] Step Six: Reassign values ​​to neurons in the special regions identified in Step Five;

[0071] Step 7: Based on the iterative 2D neural network obtained in Step 4 and the special regions reassigned in Step 6, the AUV is controlled to navigate in a "bow" shaped path in the non-special regions of the neural network. When entering a special region, the AUV's navigation is guided by the distribution of neuron activity values ​​in that region, thus completing the full-coverage path planning for the AUV. Based on the obtained full-coverage path, the full-coverage path is optimized in real time using a steering optimization strategy and a GBNN-based deadlock escape method to obtain the final navigation path of the AUV.

[0072] The process of obtaining the 2D neural network described in step three is as follows:

[0073] Based on GBNN to represent the underwater operating environment, GBNN sequentially maps the two-dimensional operating space onto a neural network. The AUV operating space contains three states of grids: grids occupied by obstacles, untraversed grids, and traversed grids. Each neuron connects only to other neurons within its eight-neighbor receptive field centered on it. The dynamic characteristics of a single neuron are described by the following equation:

[0074]

[0075] In the formula: x i (t+1) is the activity value of the i-th neuron at time t+1, x j (t) represents the activity value of the j-th neuron, which has a lateral connection with the i-th neuron, at time t. [x j (t)] + =max{x j The transfer function g(·) is defined as follows: g(t), 0}. This indicates that only positively active neurons in the GBNN can propagate globally through connections, while neurons with negative activity values ​​will be suppressed and cannot propagate through connections.

[0076]

[0077] In the formula, β is a positive constant. The transfer function g(·) ensures that the neuron activity value in the untraversed region is always 1, the neuron activity value in the traversed region is always [0,1), and the activity value in the obstacle region is -1.

[0078] Furthermore, in GBNN, the information transmission strength decreases as the distance between neurons increases. Therefore, the connection weight w between the i-th neuron and the j-th neuron... ij The definition is shown in the following formula.

[0079]

[0080] In the formula, α is a positive number, |ij| represents the Euclidean distance between the i-th neuron and the j-th neuron, and R is the receptive radius of the neuron. In this paper, the connection weights between neurons are symmetric, i.e., w ij =w ji I i It is the external input of the i-th neuron, defined by the following formula:

[0081]

[0082] In the formula, +E represents excitatory stimulation of the neuron, and -E represents inhibitory input to the neuron. It is important to note that E must be much greater than the weighted sum of the activity values ​​that have lateral connections with the i-th neuron, i.e.

[0083] The special region mentioned in step five includes the area not covered in the AUV reverse extension direction; specifically, the area not covered in the AUV reverse extension direction is: the area from the AUV to (x A ,y A When (x) A -1,y A ) and (x A +1,y A When the neuronal activity values ​​of (x) are all greater than or equal to 1, then (x) A -1,y A () is the seed point, in [1, x) A A navigable connectivity area formed within the region.

[0084] The special area mentioned in step five also includes the area containing concave obstacles in the traversal extension direction; the area containing concave obstacles in the traversal extension direction includes: the area adjacent to the AUV when it is flying straight, the area adjacent to the AUV when it turns at the obstacle boundary, and the area adjacent to the AUV when it turns at the operation boundary.

[0085] The specific area adjacent to the AUV during direct flight is:

[0086] When the AUV's heading is 90° and it is sailing to (x A ,y A When (x) A +1,y A activity value And (x) A +1,y A -1) activity value When, then respectively with (x A +1,y A ) and (x A +1,y A -1) Construct navigable connected components S for each seed point. N Connectivity S with obstaclesO ;

[0087] When the AUV's heading is -90° and it is sailing to (x A ,y A When (x) A +1,y A Neuronal activity value at ) and When, then respectively with (x A +1,y A ) and (x A +1,y A +1) Construct a navigable connected component S for the seed point. N Connectivity S with obstacles O ;

[0088] If the navigable connected region S N Contained in the connected domain S of the obstacle O Within, the navigable connected domain S N As the area immediately adjacent to the AUV during direct flight, it is recorded as a special area;

[0089] If the navigable connected region S N Not included in the connected region S of the obstacle O Within, the navigable connected domain S N It is not considered a special area.

[0090] The specific area adjacent to the obstacle boundary when the AUV turns is as follows: when the AUV is traveling at -90° to the obstacle boundary, after turning, the AUV travels to (x A ,y A ),like Then (x) A +1,y A Construct a navigable connected region with (x) as the seed point of the navigable region, and use (x) as the seed point of the navigable region. A +1,y A -1) Construct connected domains for obstacle regions based on seed points;

[0091] When the AUV is traveling at a heading of 90° and reaches the obstacle boundary, the AUV turns and travels to (x A ,y A ),like Then (x) A +1,y A Construct a navigable connected region with (x) as the seed point of the navigable region, and use (x) as the seed point of the navigable region. A +1,y A +1) constructs the obstacle connected domain for the seed point of the obstacle region;

[0092] If the navigable connected region S N Contained in the connected domain S of the obstacle OWithin, the navigable connected domain S N As the area immediately adjacent to the AUV when turning at the obstacle boundary, it is recorded as a special area;

[0093] If the navigable connected region S N Not included in the connected region S of the obstacle O Within, the navigable connected domain S N It is not considered a special area.

[0094] The specific area adjacent to the AUV when turning at the operational boundary is as follows: when the AUV is traveling at a heading of 90° or -90° to the boundary of the detection mission, the AUV turns and travels to (x A ,y A ) position, if Then (x) A +1,y A Construct a navigable connected region S from the seed point of the navigable region. N , with the yth A Column distance (x) A +1,y A The nearest obstacle grid is used as the seed point to construct the obstacle connected region S. O Extracting the nearest navigable connected domain S using edge detection methods. N Obstacles on the side boundary.

[0095] If the semi-closed region formed by the extracted obstacle boundary and the mission area boundary contains the navigable connectivity S N When this happens, navigation is possible in the connected domain S. N As the area immediately adjacent to the AUV when it turns at the work boundary, it is recorded as a special area;

[0096] If the semi-closed area formed by the extracted obstacle boundary and the mission sea area boundary does not contain the navigable connectivity S N When this happens, navigation is possible in the connected domain S. N It is not considered a special area.

[0097] The method for reassigning neuronal values ​​in the special region described in step six is ​​as follows:

[0098] First, define the gradient direction for increasing activity values; both the layer index and the neuron index in each layer increase along the gradient direction; the neuron activity value increases along the gradient direction, and the model for calculating the maximum activity value in each layer is as follows:

[0099] V(n) = 1 + η·ρ (n-1 / 2 ), n=1,2,...,N

[0100] Where: V(n) represents the maximum activity value of neurons in the nth layer; N represents the maximum number of layers in the region; η and ρ are positive numbers;

[0101] The model for calculating the remaining neuron activity in each layer is as follows:

[0102]

[0103]

[0104]

[0105]

[0106] Where: V(i,n) represents the activity value of the i-th neuron in the n-th layer, M n This indicates the number of neurons in each layer.

[0107] The path planning calculation method in step seven is as follows: After iteration, neurons in the non-special regions of the 2D neural network guide the AUV to navigate in a "bow" shaped path. When the AUV reaches a special region, it is guided by the reassigned neuron activity value distribution. The AUV navigates from the neuron with the smallest activity value to the neuron with the largest activity value in the special region, completing the navigation in the special region. During the navigation, if the AUV enters a deadlock point neuron, a GBNN-based deadlock escape method is used to help the AUV escape the deadlock point. If multiple neurons are available for navigation when the AUV turns, a turning optimization strategy is used to optimize the AUV path.

[0108] The steering optimization strategy specifically includes the following steps:

[0109] Step 1: Using the current position of the AUV (x A ,y A Based on the distribution of neuron activity values ​​within an eight-neighbor area centered on , the next navigation position (x) of the AUV is calculated. P ,y P );

[0110] Step 2: Determine (x) P ,y P +1) and (x P ,y P -1) Are all neurons unsearched? If yes, proceed to step three; otherwise, the AUV navigates to position (x). P ,y P )

[0111] Step 3: Using (x) P ,y P +1) and (x P ,y P -1) is the seed point for the navigable region in the region {(x,y)|x A =x,y A +1≤y≤ymax Constructing a connected component Ω within the} up and Ω down ;

[0112] Step 4: For Ω respectively up and Ω down The number of neurons within the turnup and turndown are counted as num_turnup and num_turndown, respectively.

[0113] Step 5: AUV prioritizes searching connected regions with fewer neurons. After the first search, the search is repeated once to search another connected region.

[0114] The GBNN-based method for escaping deadlock points specifically includes the following steps:

[0115] Step 1: Using the current position of the AUV (x A ,y A Using ) as the center, determine whether there are neurons with an activity value ≥1 in a neighborhood of radius n = 2. If yes, proceed to step three; otherwise, proceed to step two.

[0116] Step 2: Increment the value of n by one, and in (x A ,y A Using ) as the center, determine whether there are neurons with an activity value ≥1 in a neighborhood of radius n. If yes, proceed to step three; otherwise, repeat step two.

[0117] Step 3, in (x A ,y A Centered on the target, within a neighborhood of radius n, targeting unsearched neurons, the GBNN method is used to obtain an effective path from the current position of the AUV to the target position.

Claims

1. A comprehensive target search method for autonomous underwater vehicles, characterized in that, Includes the following steps: Step 1: Download the elevation map of the mission area and import it into the AUV; Step 2: Select the grid scale based on the operating radius of the detection sensor and the turning radius of the AUV, and rasterize the elevation map obtained in Step 1 to obtain a raster map; The relationship between the selected grid size L and the turning radius R1 of the AUV and the operating radius R2 of the detection sensor is as follows: Step 3: Represent the grid map obtained in Step 2 using the GBNN algorithm to obtain a 2D neural network; Step 4: Initialize and iterate the 2D neural network obtained in Step 3 to obtain the iterated 2D neural network; Step 5: Identify the navigable area affected by obstacles in the iterated 2D neural network; this navigable area is designated as a special region. Step Six: Reassign values ​​to neurons in the special regions identified in Step Five; Step 7: Based on the iterative 2D neural network obtained in Step 4 and the special regions reassigned in Step 6, the AUV is controlled to navigate in a "bow" shaped path in the non-special regions of the neural network. When entering a special region, the AUV is guided by the distribution of neuron activity values ​​in that region, thus completing the full-coverage path planning for the AUV. Based on the obtained full-coverage path, the full-coverage path is optimized in real time using a steering optimization strategy and a GBNN-based deadlock escape method to obtain the final navigation path of the AUV.

2. The method for full-coverage target search for autonomous underwater vehicles according to claim 1, characterized in that, The special region mentioned in step five includes the area not covered in the AUV reverse extension direction; specifically, the area not covered in the AUV reverse extension direction is: the area from the AUV to (x A ,y A When (x) A -1,y A ) and (x A +1,y A When the neuronal activity values ​​of (x) are all greater than or equal to 1, then (x) A -1,y A () is the seed point, in [1, x) A A navigable connectivity area formed within the region.

3. The method for full-coverage target search for autonomous underwater vehicles according to claim 2, characterized in that, The special area mentioned in step five also includes the area containing concave obstacles in the traversal extension direction; the area containing concave obstacles in the traversal extension direction includes: the area immediately adjacent when the AUV is flying straight, the area immediately adjacent when the AUV turns at the obstacle boundary, and the area immediately adjacent when the AUV turns at the operation boundary.

4. The method for full-coverage target search for autonomous underwater vehicles according to claim 3, characterized in that, The specific area adjacent to the AUV during direct flight is: When the AUV's heading is 90° and it is sailing to (x A ,y A When (x) A +1,y A activity value And (x) A +1,y A -1) activity value When, then respectively with (x A +1,y A ) and (x A +1,y A -1) Construct navigable connected regions and obstacle connected regions for seed points respectively; When the AUV's heading is -90° and it is sailing to (x A ,y A When (x) A +1,y A Neuronal activity value at ) and When, then respectively with (x A +1,y A ) and (x A +1,y A +1) Construct navigable connected regions and obstacle connected regions for the seed point; If the navigable connectivity is contained within the obstacle connectivity, the navigable connectivity is recorded as a special area as the adjacent area when the AUV is flying straight. If a navigable connected region is not contained within an obstacle connected region, then the navigable connected region is not considered a special region.

5. The method for full-coverage target search for autonomous underwater vehicles according to claim 3, characterized in that, The specific area adjacent to the obstacle boundary when the AUV turns is as follows: when the AUV is traveling at -90° to the obstacle boundary, after turning, the AUV travels to (x A ,y A ),like Then (x) A +1,y A Construct a navigable connected region with (x) as the seed point of the navigable region, and use (x) as the seed point of the navigable region. A +1,y A -1) Construct connected domains for obstacle regions based on seed points; When the AUV is traveling at a heading of 90° and reaches the boundary of the obstacle, it turns and travels to (x A ,y A ),like Then (x) A +1,y A Construct a navigable connected region with (x) as the seed point of the navigable region, and use (x) as the seed point of the navigable region. A +1,y A +1) constructs the obstacle connected domain for the seed point of the obstacle region; If the navigable connected region is contained within the obstacle connected region, the navigable connected region is recorded as a special region as the area immediately adjacent to the AUV when it turns at the obstacle boundary. If a navigable connected region is not contained within an obstacle connected region, then the navigable connected region is not considered a special region.

6. The method for full-coverage target search for autonomous underwater vehicles according to claim 3, characterized in that, The specific area adjacent to the AUV when turning at the operational boundary is as follows: when the AUV is traveling at a heading of 90° or -90° to the boundary of the detection mission, the AUV turns and travels to (x... A ,y A ) position, if Then (x) A +1,y A Construct a navigable connected region using the seed point of the navigable region as the y-th point. A Column distance (x) A +1,y A The nearest obstacle grid is used as the seed point to construct the obstacle connected domain; the obstacle boundary on the side of the adjacent navigable connected domain is extracted by the edge detection method; If the semi-closed area formed by the extracted obstacle boundary and the mission sea area boundary contains a navigable connected area, then the navigable connected area is recorded as a special area as the area adjacent to the AUV when it turns at the operational boundary. If the semi-closed area formed by the extracted obstacle boundary and the mission sea area boundary does not contain a navigable connected domain, then the navigable connected domain is not considered a special area.

7. The method for full-coverage target search for autonomous underwater vehicles according to claim 1, characterized in that, The method for reassigning neuronal values ​​in the special region described in step six is ​​as follows: First, define the gradient direction for increasing activity values; both the layer index and the neuron index in each layer increase along the gradient direction; the neuron activity value increases along the gradient direction, and the model for calculating the maximum activity value in each layer is as follows: V(n)=1+n·p (n-1 / 2) ,n=1,2,...,N Where: V(n) represents the maximum activity value of neurons in the nth layer; N represents the maximum number of layers in the region; η and ρ are positive numbers; The model for calculating the remaining neuron activity in each layer is as follows: Where: V(i,n) represents the activity value of the i-th neuron in the n-th layer, M n This indicates the number of neurons in each layer.

8. The method for full-coverage target search for autonomous underwater vehicles according to claim 1, characterized in that, The path planning calculation method in step seven is as follows: After iteration, neurons in the non-special regions of the 2D neural network guide the AUV to navigate in a "bow" shaped path. When the AUV navigates to a special region, it is guided by the reassigned neuron activity value distribution. The AUV navigates from the neuron with the smallest activity value to the neuron with the largest activity value in the special region, completing the navigation in the special region. During the navigation, if the AUV enters a deadlock point neuron, a deadlock escape method based on GBNN is used to help the AUV escape the deadlock point. If multiple neurons are available for navigation when the AUV turns, a turning optimization strategy is used to optimize the AUV path.

9. A method for full-coverage target search for autonomous underwater vehicles according to claim 1, characterized in that, The steering optimization strategy specifically includes the following steps: Step 1: Using the current position of the AUV (x A ,y A Based on the distribution of neuron activity values ​​within an eight-neighbor area centered on , the next navigation position (x) of the AUV is calculated. P ,y P ); Step 2: Determine (x) P ,y P +1) and (x P ,y P -1) Are all neurons unsearched? If yes, proceed to step three; otherwise, the AUV navigates to position (x). P ,y P ) Step 3: Using (x) P ,y P +1) and (x P ,y P -1) is the seed point for the navigable region in the region {(x,y)|x A =x,y A +1≤y≤y max Constructing a connected region Ω within the} up and Ω down ; Step 4: For Ω respectively up and Ω down The number of neurons within the turnup and turndown are counted as num_turnup and num_turndown, respectively. Step 5: AUV prioritizes searching connected regions with fewer neurons. After the first search, the search is repeated once to search another connected region.

10. A method for full-coverage target search for autonomous underwater vehicles according to claim 9, characterized in that, The GBNN-based method for escaping deadlock points specifically includes the following steps: Step 1: Using the current position of the AUV (x A ,y A Using ) as the center, determine whether there are neurons with an activity value ≥1 in a neighborhood of radius n = 2. If yes, proceed to step three; otherwise, proceed to step two. Step 2: Increment the value of n by one, and in (x A ,y A Using ) as the center, determine whether there are neurons with an activity value ≥1 in a neighborhood of radius n. If yes, proceed to step three; otherwise, repeat step two. Step 3, in (x A ,y A Centered on the target, within a neighborhood of radius n, targeting unsearched neurons, the GBNN method is used to obtain an effective path from the current position of the AUV to the target position.

Images