Marine bamboo raft target extraction method and system

Abstract

The invention belongs to the field of computer vision, image processing, neurodynamics algorithms and control, and discloses an offshore bamboo raft target extraction method and system, and the method comprises the steps: constructing an ocean target extraction model; converting the ocean target extraction model with constraints into an unconstrained optimization model by using a penalty function; establishing a collaborative integral enhanced gradient descent neurodynamics algorithm to solve the ocean target extraction model; introducing a particle swarm algorithm to assist in quickly searching a filtering vector; according to the collaborative integral enhanced gradient descent neurodynamics algorithm, gradient descent calculation is carried out at each position through randomly scattering particles, then judgment is carried out according to a current optimal solution and a global optimal solution, an optimal filtering vector is obtained, and the optimal filtering vector is calculated. The efficiency and the accuracy of the gradient descent neurodynamics algorithm for offshore bamboo raft target extraction are greatly improved; and finally, an optimal extraction threshold value is obtained by using an Otus algorithm, and the extraction of the offshore bamboo raft target is completed.

Description

Technical Field

[0001] The invention belongs to the fields of computer vision, image processing, neural dynamics algorithm and control, and specifically relates to a method and system for extracting a sea bamboo raft target. Background Art

[0002] With the growing demand for marine resource development and fishery supervision, marine target detection technology based on remote sensing images or visible light photography is of great significance in environmental monitoring, illegal fishing control, maritime search and rescue, etc. Among them, bamboo rafts, as a typical unstructured small target, have become a technical difficulty in marine target detection due to their small size, low contrast between the material and the seawater background, and susceptibility to wave and light interference.

[0003] Traditional marine target detection algorithms are mostly based on threshold segmentation, edge detection or texture analysis (such as gray-level co-occurrence matrix). Such methods are easily affected by wave texture, reflection interference and cloud noise in complex sea conditions, resulting in significantly increased false detection and missed detection rates. For example, segmentation algorithms based on fixed thresholds are difficult to adapt to sea scenes under dynamic lighting changes, and traditional morphological methods have limited ability to distinguish between bamboo rafts and floating objects (such as fishing nets and plastic waste). In recent years, although deep learning-based detection algorithms (such as Faster R-CNN and YOLO series) have performed well in general target detection, they still have the following problems when directly applied to bamboo raft detection:

[0004] 1. Loss of small target features: Bamboo rafts usually only occupy dozens of pixels in remote sensing images. The downsampling operation of the existing network easily leads to the loss of shallow features, making it difficult to capture the detailed information of the bamboo raft;

[0005] 2. Sensitive to background interference: The similarity of local textures of wave ripples, cloud shadows, and bamboo rafts can easily cause false alarms, while the existing attention mechanism is insufficient in modeling long-range dependencies and is difficult to effectively suppress background noise;

[0006] 3. Insufficient real-time performance: Although lightweight models (such as MobileNet) can improve detection speed, their accuracy drops sharply in complex sea scenes, making it difficult to strike a balance between efficiency and accuracy;

[0007] 4. High data dependence: There is a lack of multi-angle and multi-scale labeled samples for bamboo rafts in the public dataset, which limits the generalization ability of the model, especially in low light and rainy and foggy weather, where the performance degrades significantly.

[0008] Therefore, there is an urgent need for a dedicated target extraction algorithm tailored to the characteristics of offshore bamboo rafts, which can achieve high-precision and high-efficiency detection under complex sea conditions, while reducing dependence on labeled data to meet the needs of actual application scenarios. Summary of the invention

[0009] In order to solve the problems existing in the prior art, the present invention provides a method and system for extracting a bamboo raft target at sea, aiming to improve the efficiency and accuracy of computer extraction of target objects.

[0010] To achieve the above object, the present invention provides the following solutions:

[0011] A method for extracting a target from a bamboo raft at sea, the method comprising:

[0012] S1. Construct a marine target extraction model based on the constrained energy minimization scheme and sparse theory;

[0013] S2, using penalty function and Lagrange number multiplication, the ocean target extraction model is converted into an unconstrained optimization model;

[0014] S3, calculate the gradient of the unconstrained optimization model at the current moment;

[0015] S4. Introduce the integral enhanced gradient descent neural dynamics algorithm, input each state variable of the ocean target extraction model at the current moment into the integral enhanced gradient descent neural dynamics algorithm, construct the integral enhancement term based on the calculated gradient, calculate the local optimal solution under the current state, and pass it to the next moment;

[0016] S5. Introduce the particle swarm algorithm, record the local optimal solution at each location, and select the best one as the global optimal solution to make the next step;

[0017] S6. Update each state variable of the constructed ocean target extraction model and return to S4 until the current number of iterations exceeds the preset value, or the algorithm error is lower than the preset value, obtain the spectral information matrix of the target extraction, and complete the extraction of the sea raft target.

[0018] Preferably, in S1, constructing an ocean target extraction model based on a constrained energy minimization scheme and sparse theory includes:

[0019] Assume that any remote sensing image is in matrix form Indicates that Represents the total number of pixels, Indicates the number of bands, Represents remote sensing image pixels, and based on the sparse theory, a model for extracting ocean targets is constructed:

[0020] ;

[0021] in, represents the autocorrelation matrix, represents the target vector, represents the filter vector.

[0022] Preferably, in S2, using a penalty function and Lagrange number multiplication to convert the ocean target extraction model into an unconstrained optimization model comprises:

[0023] .

[0024] Preferably, in S3, calculating the gradient of the unconstrained optimization model at the current moment includes:

[0025] Assume that the main function of the unconstrained optimization model is , based on the main function of the unconstrained optimization model, calculate the gradient .

[0026] Preferably, in S4, introducing the integral enhanced gradient descent neural dynamics algorithm comprises:

[0027] Construct an integral enhancement term based on the computed gradient:

[0028] ;

[0029] in, Indicates the extraction time. is the integration variable, Express The differential of

[0030] Based on the integral enhancement term, an integral enhanced gradient descent neural dynamics algorithm is established:

[0031] ;

[0032] in, is the scale factor, is the integration coefficient, express right Derivation;

[0033] The discrete form is expressed as:

[0034] ;

[0035] in, Indicates Iterations, Indicates The filter vector in iteration .

[0036] Preferably, in S5, introducing a particle swarm algorithm, recording the local optimal solution at each position, and judging and selecting the best one as the global optimal solution includes:

[0037] ;

[0038] in, Indicates The speed of a particle, Indicates The position of a particle, , A random number, It is The particle in The best historical position in iterations, It is The global best position in iteration .

[0039] Preferably, in S6, each state variable of the constructed ocean target extraction model is updated, and the process returns to S4 until the current number of iterations exceeds a preset value, or the algorithm error is lower than a preset value, and the spectral information matrix of the target extraction is obtained, and the extraction of the sea raft target is completed, including:

[0040] S601: Preset number of particles ; Preset maximum number of iterations ; Preset inertia weight ; Preset individual learning factor ; Preset social learning factor ; Preset autocorrelation matrix ; Preset initial filter vector ;

[0041] S602: Calculate the gradient of the unconstrained optimization model at the current iteration, that is, , if the number of iterations If it is greater than the maximum number of iterations, the calculation is stopped and the optimal solution is output; otherwise, the process goes to step S603;

[0042] S603: Establish an integral enhanced gradient descent neural dynamics algorithm to calculate , obtain the local optimal solution;

[0043] S604: According to the particle swarm algorithm, the initial values ​​are randomly scattered at various positions, and the integral enhanced gradient descent neural dynamics algorithm is used at each position to search for a local optimal solution;

[0044] S605: In each search, record the optimal solution of each position, and then find the best vector therein as the current local optimal vector;

[0045] S606: In each iteration process, compare the recorded local optimal vectors and select the best one as the global optimal vector;

[0046] S607: Search rules based on particle swarm , update the search position of the vector;

[0047] S608: Go to step S602;

[0048] S609: Obtain the spectral information matrix of target extraction and complete the extraction of the sea raft target.

[0049] The present invention also provides a marine bamboo raft target extraction system, the system is used to implement the above method, the system comprises: a construction module, a conversion module, a calculation module, a local optimal solution module, a global optimal solution module and an extraction module;

[0050] The building module is used to build an ocean target extraction model based on a constrained energy minimization scheme and sparse theory;

[0051] The conversion module is used to convert the ocean target extraction model into an unconstrained optimization model by using a penalty function and Lagrange number multiplication;

[0052] The calculation module is used to calculate the gradient of the unconstrained optimization model at the current moment;

[0053] The local optimal solution module is used to introduce an integral enhanced gradient descent neural dynamics algorithm, input each state variable of the ocean target extraction model at the current moment into the integral enhanced gradient descent neural dynamics algorithm, construct an integral enhancement function based on the calculated gradient, calculate the local optimal solution under the current state, and pass it to the next moment;

[0054] The global optimal solution module is used to introduce the particle swarm algorithm, record the local optimal solution at each location, and determine and select the best one as the global optimal solution, so as to make the next step;

[0055] The extraction module is used to update the state variables of the constructed ocean target extraction model and return to the local optimal solution module until the current number of iterations exceeds a preset value or the algorithm error is lower than a preset value, thereby obtaining the spectral information matrix of the target extraction and completing the extraction of the offshore bamboo raft target.

[0056] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the aforementioned method when executing the program.

[0057] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed, the aforementioned method is implemented.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] (1) The present invention is based on the extraction method of the collaborative integral enhanced gradient descent neural dynamics algorithm, which calculates the gradient of the unconstrained optimization model , forming the local position iteration direction and step size;

[0060] (2) This invention is inspired by control theory and is based on calculating the gradient Reasonable construction of integral enhancement terms makes the collaborative integral enhancement gradient descent neural dynamics algorithm have strong noise tolerance;

[0061] (3) Inspired by the particle swarm algorithm, the present invention randomly scatters , and perform local solution based on the integral enhanced gradient descent neural dynamics algorithm to obtain the global optimal solution at the current iteration moment;

[0062] (4) The method of the present invention uses the integral enhanced gradient descent neural dynamics algorithm to calculate the filter vector extracted at the next moment at each iteration; and in each solution process, all information is summarized to obtain a local optimal solution at the current iteration moment, and finally a global optimal solution is obtained;

[0063] (5) The method of the present invention makes full use of the particle swarm algorithm in the solution process, has a strong global search capability, improves the accuracy of the integral enhanced gradient descent neural dynamics algorithm, and thus improves the accuracy of computer vision target extraction.

[0064] (6) At the same time, the method of the present invention solves the image spectral information based on the collaborative integral enhanced gradient descent neural dynamics algorithm, avoiding the pseudo-inverse of the relationship matrix between the autocorrelation coefficient matrix and the actual image spectral information, and can improve the computer's solution efficiency for target extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0066] Figure 1 A flow chart of a method for extracting a sea bamboo raft target in an embodiment of the present invention;

[0067] Figure 2 is the original image in the embodiment of the present invention, and the image includes the bamboo raft on the sea to be extracted this time;

[0068] Figure 3 is a true value image of the original image in the embodiment of the present invention, wherein the white part is the bamboo raft on the sea to be extracted;

[0069] Figure 4 It is the result extracted by using the collaborative integral enhanced gradient descent neural dynamics algorithm proposed in the embodiment of the present invention;

[0070] Figure 5 This is the result extracted by using the particle swarm algorithm in the embodiment of the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of 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.

[0072] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0073] Embodiment 1

[0074] like Figure 1 As shown, an embodiment of the present invention discloses a method for extracting a bamboo raft target at sea, the method comprising:

[0075] S1. Construct a marine target extraction model based on the constrained energy minimization scheme and sparse theory;

[0076] S2, using penalty function and Lagrange number multiplication, the ocean target extraction model is converted into an unconstrained optimization model;

[0077] S3, calculate the gradient of the unconstrained optimization model at the current moment;

[0078] S4. Introduce the integral enhanced gradient descent neural dynamics algorithm, input each state variable of the ocean target extraction model at the current moment into the integral enhanced gradient descent neural dynamics algorithm, construct an integral enhancement function based on the calculated gradient, calculate the local optimal solution under the current state, and pass it to the next moment;

[0079] S5. Introduce the particle swarm algorithm, record the local optimal solution at each location, and select the best one as the global optimal solution to make the next step;

[0080] S6. Update each state variable of the constructed ocean target extraction model and return to S4 until the current number of iterations exceeds the preset value, or the algorithm error is lower than the preset value, obtain the spectral information matrix of the target extraction, and complete the extraction of the sea raft target.

[0081] In this embodiment, S1, constructing an ocean target extraction model:

[0082] Assume that any remote sensing image can be expressed in matrix form Indicates that Represents the total number of pixels, Indicates the number of bands, Represents remote sensing image pixels, and based on the sparse theory, the following ocean target extraction model can be constructed:

[0083] ;

[0084] in, represents the autocorrelation matrix, represents the target vector, represents the filter vector.

[0085] In this embodiment, S2, the ocean target extraction model is converted into an unconstrained optimization model using a penalty function:

[0086] Specifically, according to the penalty function, the ocean target extraction model can be converted into an unconstrained optimization model:

[0087] ;

[0088] In this embodiment, S3, calculate the gradient of the unconstrained optimization model at the current moment:

[0089] Specifically, let’s assume that the main function of the unconstrained optimization model is ,

[0090] Then we ask: .

[0091] In this embodiment, S4 introduces an integral-enhanced gradient descent neural dynamics algorithm, inputs each state variable of the ocean target extraction model at the current moment into the integral-enhanced gradient descent neural dynamics algorithm, constructs an integral enhancement term based on the calculated gradient, calculates the local optimal solution in the current state, and passes it to the next moment.

[0092] Specifically, we first construct an integral enhancement term to adapt to the complex ocean environment.

[0093] ;

[0094] in, Indicates the extraction time. is the integration variable, Express The differential of .

[0095] Then, the integral enhanced gradient descent neural dynamics algorithm is established

[0096] ;

[0097] in, is the scale factor, is the integration coefficient, express right Derivative.

[0098] Specifically, its discrete form can be expressed as

[0099] ;

[0100] in, Indicates Iterations, Indicates The filter vector in iteration .

[0101] According to the maximum number of iterations set, the filter vector in this case is calculated , which is the local optimal solution in the current state.

[0102] In this embodiment, S5, a particle swarm algorithm is introduced to record the local optimal solution at each position, and the best one is selected as the global optimal solution, so as to make the next step:

[0103] Specifically, the search rules of the particle swarm are:

[0104] ;

[0105] in, Indicates The speed of a particle, Indicates The position of a particle, , A random number, It is The particle in The best historical position in iterations, It is The global best position in iteration .

[0106] In this embodiment, S6 updates the state variables of the constructed computer vision target extraction model and returns to S4 until the current number of iterations exceeds the preset value or the algorithm error is lower than the preset value, and obtains the spectral information matrix of the target extraction to perform subsequent target extraction tasks.

[0107] The specific operation includes the following steps:

[0108] S601: parameter initialization;

[0109] Given number of particles ; Given a maximum number of iterations ; Given inertia weight ; Given individual learning factor ; Given the social learning factor ;

[0110] Given the autocorrelation matrix ; Given the initial filter vector ;

[0111] S602: Calculate the gradient of the unconstrained optimization model at the current iteration, that is, , if the number of iterations If it is greater than the maximum number of iterations, the calculation is stopped and the optimal solution is output; otherwise, the process goes to step S603;

[0112] S603: Design of an integral enhanced gradient descent neural dynamics algorithm, calculation

[0113] ;

[0114] Obtain local optimal solution;

[0115] S604: According to the particle swarm algorithm, the initial values ​​are randomly scattered at various positions, and the integral enhanced gradient descent neural dynamics algorithm is used at these positions to search for local optimal solutions;

[0116] S605: In each search, record the optimal solution of each position, and then find the best vector therein as the current local optimal vector;

[0117] S606: In each iteration process, compare the recorded local optimal vectors and select the best one as the global optimal vector;

[0118] S607: Search rules based on particle swarm, which are set as

[0119] ;

[0120] Update the search position of the vector;

[0121] S608: Go to step S602.

[0122] The target extraction method based on the collaborative integral enhanced gradient descent neural algorithm of the present invention is further described below with specific embodiments.

[0123] Specifically, the method of the present invention is experimentally simulated by taking the extraction of a target object in an image as an example using MATLAB software and ENVI software.

[0124] Figure 2 This is the original image in this example, which contains the bamboo raft on the sea to be extracted.

[0125] Figure 3 It is the true value image of the original image, where the white part is the bamboo raft on the sea to be extracted.

[0126] Figure 4 This is the result extracted using the collaborative integral enhanced gradient neural dynamics algorithm proposed in the present invention.

[0127] Figure 5 The results are extracted using the particle swarm algorithm.

[0128] Table 1 Experiments on extracting bamboo rafts at sea using different algorithms in satellite remote sensing images, with evaluation indicators: overall accuracy, precision, recall rate, Kappa coefficient and F1-score.

[0129] Table 1

[0130]

[0131] from Figure 4 and Figure 5 From the above, the proposed collaborative integral enhanced gradient descent neural dynamics algorithm has better extraction effect than the simple particle swarm algorithm with fewer iterations and particles. The collaborative integral enhanced gradient descent neural dynamics algorithm proposed in the present invention has the ability to quickly extract target objects, can ensure strong local search capabilities and high noise tolerance, and incorporates the global search capabilities of the particle swarm algorithm to achieve fast and accurate target extraction. Figure 5 As shown in Figure 3), it is not possible to quickly and accurately search for the optimal filter vector, resulting in a weak extraction effect.

[0132] In addition, Table 1 shows the evaluation indicators of the sea bamboo raft extraction experiment using different algorithms in satellite remote sensing images. There is no doubt that the collaborative integral enhanced gradient descent neural dynamics algorithm proposed in the present invention has outstanding advantages, each of which is higher than the particle swarm algorithm. The Kappa coefficient and F1-score of the collaborative integral enhanced gradient descent neural dynamics algorithm are both higher than 0.88, which is enough to prove its excellent extraction ability. In addition, the overall accuracy reached above 0.99. The Kappa coefficient and F1-score of the particle swarm algorithm are both below 0.03, and the target object is basically unable to be extracted. This shows that the proposed collaborative integral enhanced gradient descent neural dynamics algorithm has excellent performance in the extraction of sea bamboo raft targets under complex backgrounds, greatly improving the efficiency and accuracy of collaborative neural dynamics for computer vision target extraction.

[0133] Embodiment 2

[0134] The present invention also discloses a marine bamboo raft target extraction system, the system is used to implement the method described in the first embodiment, the system comprises: a construction module, a conversion module, a calculation module, a local optimal solution module, a global optimal solution module and an extraction module;

[0135] A building block for constructing a marine object extraction model based on a constrained energy minimization scheme and sparsity theory;

[0136] A conversion module, used for converting the ocean target extraction model into an unconstrained optimization model by using a penalty function and Lagrangian number multiplication;

[0137] The calculation module is used to calculate the gradient of the unconstrained optimization model at the current moment;

[0138] The local optimal solution module is used to introduce the integral enhanced gradient descent neural dynamics algorithm, input each state variable of the ocean target extraction model at the current moment into the integral enhanced gradient descent neural dynamics algorithm, construct an integral enhancement function based on the calculated gradient, calculate the local optimal solution under the current state, and pass it to the next moment;

[0139] The global optimal solution module is used to introduce the particle swarm algorithm, record the local optimal solution at each location, and select the best one as the global optimal solution to make the next step;

[0140] The extraction module is used to update the state variables of the constructed marine target extraction model and return to the local optimal solution module until the current number of iterations exceeds the preset value or the algorithm error is lower than the preset value, so as to obtain the spectral information matrix of the target extraction and complete the extraction of the marine bamboo raft target.

[0141] Embodiment 3

[0142] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the offshore bamboo raft target extraction method in the above-mentioned embodiment when executing the program.

[0143] Embodiment 4

[0144] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed, the offshore bamboo raft target extraction method in the above embodiment is implemented.

[0145] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for extracting a target from a bamboo raft at sea, characterized in that: The method comprises: S1. Construct a marine target extraction model based on the constrained energy minimization scheme and sparse theory; S2, using penalty function and Lagrange number multiplication, the ocean target extraction model is converted into an unconstrained optimization model; S3, calculate the gradient of the unconstrained optimization model at the current moment; S4. Introduce the integral enhanced gradient descent neural dynamics algorithm, input each state variable of the ocean target extraction model at the current moment into the integral enhanced gradient descent neural dynamics algorithm, construct the integral enhancement term based on the calculated gradient, calculate the local optimal solution under the current state, and pass it to the next moment; S5. Introduce the particle swarm algorithm, record the local optimal solution at each location, and select the best one as the global optimal solution to make the next step; S6. Update each state variable of the constructed ocean target extraction model and return to S4 until the current number of iterations exceeds the preset value, or the algorithm error is lower than the preset value, obtain the spectral information matrix of the target extraction, and complete the extraction of the sea raft target.

2. The method according to claim 1, characterized in that In S1, based on the constrained energy minimization scheme and sparse theory, the ocean target extraction model is constructed including: Assume that any remote sensing image is in matrix form Indicates that Represents the total number of pixels, Indicates the number of bands, Represents remote sensing image pixels, and based on the sparse theory, a model for extracting ocean targets is constructed: ； in, represents the autocorrelation matrix, represents the target vector, represents the filter vector.

3. The method according to claim 2, characterized in that In S2, using the penalty function and Lagrange number multiplication, the ocean target extraction model is converted into an unconstrained optimization model, including: 。 4. The method according to claim 3, characterized in that In S3, calculating the gradient of the unconstrained optimization model at the current moment includes: Assume that the main function of the unconstrained optimization model is , based on the main function of the unconstrained optimization model, calculate the gradient .

5. The method according to claim 4, characterized in that In S4, the gradient descent neural dynamics algorithm with integral enhancement is introduced, including: Construct an integral enhancement term based on the computed gradient: ； in, Indicates the extraction time. is the integration variable, Express The differential of Based on the integral enhancement term, an integral enhanced gradient descent neural dynamics algorithm is established: ； in, is the scale factor, is the integration coefficient, express right Derivation; The discrete form is expressed as: ； in, Indicates Iterations, Indicates The filter vector in iteration .

6. The method according to claim 5, characterized in that In S5, a particle swarm algorithm is introduced to record the local optimal solution at each position, and the best one is selected as the global optimal solution, including: ； in, Indicates The speed of a particle, Indicates The position of a particle, , A random number, It is The particle in The best historical position in iterations, It is The global best position in iteration .

7. The method according to claim 6, characterized in that In the S6, each state variable of the constructed ocean target extraction model is updated, and the process returns to S4 until the current number of iterations exceeds a preset value, or the algorithm error is lower than a preset value, and the spectral information matrix of the target extraction is obtained. The extraction of the sea raft target is completed, including: S601: Preset number of particles ; Preset maximum number of iterations ; Preset inertia weight ; Preset individual learning factor ; Preset social learning factor ; Preset autocorrelation matrix ; Preset initial filter vector ; S602: Calculate the gradient of the unconstrained optimization model at the current iteration, that is, , if the number of iterations If it is greater than the maximum number of iterations, the calculation is stopped and the optimal solution is output; otherwise, the process goes to step S603; S603: Establish an integral enhanced gradient descent neural dynamics algorithm to calculate , obtain the local optimal solution; S604: According to the particle swarm algorithm, the initial values ​​are randomly scattered at various positions, and the integral enhanced gradient descent neural dynamics algorithm is used at each position to search for a local optimal solution; S605: In each search, the optimal solution of each position is recorded, and then the best vector is found as the current local optimal vector; S606: In each iteration process, compare the recorded local optimal vectors and select the best one as the global optimal vector; S607: Search rules based on particle swarm , update the search position of the vector; S608: Go to step S602; S609: Obtain the spectral information matrix of target extraction and complete the extraction of the sea raft target.

8. A marine bamboo raft target extraction system, the system being used to implement the method described in any one of claims 1 to 7, characterized in that: The system comprises: a construction module, a conversion module, a calculation module, a local optimal solution module, a global optimal solution module and an extraction module; The building module is used to build an ocean target extraction model based on a constrained energy minimization scheme and sparse theory; The conversion module is used to convert the ocean target extraction model into an unconstrained optimization model by using a penalty function and Lagrange number multiplication; The calculation module is used to calculate the gradient of the unconstrained optimization model at the current moment; The local optimal solution module is used to introduce an integral enhanced gradient descent neural dynamics algorithm, input each state variable of the ocean target extraction model at the current moment into the integral enhanced gradient descent neural dynamics algorithm, construct an integral enhancement function based on the calculated gradient, calculate the local optimal solution under the current state, and pass it to the next moment; The global optimal solution module is used to introduce the particle swarm algorithm, record the local optimal solution at each location, and determine and select the best one as the global optimal solution, so as to make the next step; The extraction module is used to update the state variables of the constructed ocean target extraction model and return to the local optimal solution module until the current number of iterations exceeds a preset value or the algorithm error is lower than a preset value, thereby obtaining the spectral information matrix of the target extraction and completing the extraction of the offshore bamboo raft target.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images