A method for detecting aquaculture cages based on adaptive variable parameter dynamic learning networks

Abstract

This invention discloses a method for detecting aquaculture cages in remote sensing images based on an adaptive variable parameter dynamic learning network. The method includes: first, calculating the autocorrelation matrix of the input raw remote sensing image and establishing an equation-constrained optimization mathematical model for its filtered output; then, transforming the linear equation mathematical model using the Lagrange multiplier method and solving it using an adaptive variable parameter dynamic learning network algorithm; finally, passing the output filter coefficient vector through a transformation domain to obtain the filtered remote sensing image, thus achieving the detection of aquaculture cages in remote sensing images. The adaptive variable parameter dynamic learning network proposed in this invention performs excellently in image classification and target detection, not only improving the accuracy of target detection in remote sensing images but also shortening the computation time and enhancing target classification capabilities.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology and image processing technology, and particularly relates to a remote sensing image aquaculture cage detection method based on an adaptive variable parameter dynamic learning network. Background Technology

[0002] In recent years, the scale of nearshore aquaculture in China has continued to expand, with a rich variety of aquaculture species. Various aquaculture methods, primarily cage culture, have been adopted, and technological advancements have significantly improved aquaculture efficiency and product quality. However, because cage culture is mostly conducted in shallow waters less than 20 meters deep, the relatively enclosed terrain and poor exchange with the open sea mean that feed residues, fish feces, and drug residues can lead to eutrophication, algal blooms, and other environmental problems. Furthermore, some sea areas suffer from overly dense cage layouts and a lack of scientific planning, further exacerbating environmental issues. Therefore, precise monitoring of cage layouts and surrounding waters is necessary. Traditional manual inspection methods, due to insufficient coverage and efficiency, cannot meet the monitoring needs of modern large-scale precision aquaculture. Therefore, the introduction of remote sensing technology has become an inevitable trend.

[0003] Remote sensing acquires environmental information via satellites or aircraft, offering advantages such as wide coverage, high real-time performance, and low cost. It can be used for efficient monitoring and management of aquaculture areas, promoting the sustainable development of the aquaculture industry. Remote sensing technology overcomes many shortcomings of traditional monitoring methods, possessing greater spatial freedom and development potential. With the continuous upgrading of satellite remote sensing sensors, hyperspectral and high spatial resolution remote sensing imagery technologies have rapidly developed, making land cover classification and extraction a research focus. From early visual interpretation to classic classification methods, and then to neural networks and deep learning algorithms, remote sensing extraction of aquaculture areas has achieved significant results. Effective extraction typically relies on the identification and segmentation of remote sensing image data, and depends on intelligent algorithms, including pixel classification, object-oriented methods, and neural networks. Due to the complex background and the lack and uncertainty of information in remote sensing images, the design of target detection algorithms faces significant challenges. This complex background often makes it difficult to obtain accurate and reliable information; therefore, most target extraction methods rely on easily obtainable and highly accurate prior target spectral information. However, neglecting the utilization of background spectral information may limit further improvements in detection accuracy and reliability. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a remote sensing image-based method for detecting aquaculture cages using an adaptive variable parameter dynamic learning network, thereby resolving the issues present in the existing technologies.

[0005] To achieve the above objectives, this invention provides a remote sensing image aquaculture cage detection method based on an adaptive variable parameter dynamic learning network, comprising:

[0006] Obtain the original remote sensing image and construct the autocorrelation matrix of the original remote sensing image;

[0007] Based on the autocorrelation matrix, construct an equation-constrained optimization mathematical model;

[0008] Based on the aforementioned equality constraint optimization mathematical model, a linear equality equation mathematical model is obtained;

[0009] An adaptive variable parameter dynamic learning network is used to solve the mathematical model of the linear equation, and the filtered remote sensing image is obtained based on the solution result to realize the detection of aquaculture cages in remote sensing images.

[0010] Optionally, a matrix representation of the original remote sensing image is obtained, and an autocorrelation matrix of the original remote sensing image is obtained based on the matrix representation; wherein the matrix representation is as follows:

[0011] ;

[0012] In the formula, It refers to the number of sensors. For a picture of size A grayscale image.

[0013] Optionally, the mathematical model for optimizing the equality constraints is:

[0014] ;

[0015] in, These are the filter coefficients; It is the transpose operator; Let A be the target spectral vector and A be the autocorrelation matrix.

[0016] Optionally, the equation-constrained optimization mathematical model is transformed into an unconstrained optimization mathematical model using the Lagrange multiplier method, and a linear equation mathematical model is obtained based on the unconstrained optimization mathematical model; wherein, the linear equation mathematical model is as follows:

[0017] ;

[0018] In the formula, the autocorrelation coefficient matrix Coefficient vector ; Let be the vector to be solved; It is composed of filter coefficients. dimensional vector; It is a Lagrange multiplier.

[0019] Optionally, the unconstrained optimization model is:

[0020]

[0021] in, It is a Lagrange multiplier.

[0022] Optionally, the process of solving the mathematical model of the linear equation using an adaptive variable parameter dynamic learning network includes:

[0023] The parameters are initialized and an error function is constructed. The error function is calculated and iterated using adaptively changing parameters. It is determined whether the error function satisfies the loop condition. If the loop termination condition is met, the solution vector is output. If not, the adaptively changing parameters are updated and the error function is calculated again until the loop termination condition is met.

[0024] Optionally, the process of obtaining the filtered remote sensing image includes:

[0025] The vector to be solved is decomposed to obtain the filter coefficient vector; the filter coefficient vector is then transformed into the transform domain to obtain the filtered remote sensing image.

[0026] The present invention also discloses a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0027] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0028] The present invention also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0029] Compared with the prior art, the present invention has the following advantages and technical effects:

[0030] 1) The network error function can be in matrix, vector, or scalar form depending on the problem type, and may be an unbounded real or complex number. By combining the error function with the evolution formula, the computational error of each element can be made to converge to zero;

[0031] 2) At the system level, the derivative information of the error function is used to make the system eventually evolve towards zero error and tend to stabilize;

[0032] 3) It introduces an adaptive variable parameter in the design model. As the error function converges, the adaptive variable parameter will adaptively adjust the parameter change, gradually amplifying and shortening the exponential convergence time, enabling it to achieve fast convergence and have noise resistance. Attached Figure Description

[0033] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0034] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0035] Figure 2 The original remote sensing image is from an embodiment of the present invention.

[0036] Figure 3 The image shown is an example of aquaculture cage target detection processed by the method described in this embodiment of the invention. Detailed Implementation

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0038] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0039] Example 1

[0040] like Figure 1-3 As shown, this embodiment provides a remote sensing image aquaculture cage detection method based on an adaptive variable parameter dynamic learning network, including:

[0041] Step 1: Any remote sensing image can be represented in matrix form as follows ,in It refers to the number of sensors. For a picture of size A grayscale image.

[0042] Step 2: Based on the characteristics of the remote sensing image, the autocorrelation matrix can be calculated. .

[0043] Step 3: Next, design an optimization mathematical model with equality constraints.

[0044] (1)

[0045] These are the filter coefficients; It is the transpose operator; The target spectral vector.

[0046] Step 4: Use the Lagrange multiplier method to transform the equation-constrained optimization mathematical model (1) from the filtered output in Step 3 into an unconstrained optimization mathematical model, as shown below:

[0047] (2)

[0048] in, It is a Lagrange multiplier.

[0049] Step 5: Transform the unconstrained optimization mathematical model (2) from Step 4 into a linear equation mathematical model, as shown below:

[0050] (3)

[0051] Among them, the autocorrelation coefficient matrix Coefficient vector ; Let be the vector to be solved; It is composed of filter coefficients. dimensional vector; It is a Lagrange multiplier.

[0052] Step 6: By sampling the mathematical model (3) of the linear equation from Step 5, its corresponding discrete-time expression is obtained. Then, the mathematical model (3) of the linear equation from Step 5 is solved using an adaptive variable parameter dynamic learning network. The specific algorithm steps are as follows:

[0053] 6-1) Parameter initialization;

[0054] 6-1-1) Initialize allowable error ;

[0055] 6-1-2) Randomly generate initial points ;

[0056] 6-1-3) Initialize the step size factor Initialize the autocorrelation coefficient matrix. Initialize the coefficient vector Initialize the adaptive parameter expression. Initialize the number of iterations. ,make ;

[0057] 6-2) Calculate the error function ,like Stop calculation and output. ;

[0058] 6-3) Output based on adaptively changing parameters. ;in ,in This is the updated error function. It is achieved through adaptive parameter changes. Its function is to adaptively change parameters. Will follow The convergence of the time gradually amplifies and then shortens the exponential convergence time.

[0059] 6-4) Order . Physically, this can be understood as the direction of error adjustment, which is achieved through the autocorrelation matrix. Errors are weighted to control the step size and direction of parameter updates.

[0060] 6-5) Order , Proceed to step 6-2).

[0061] Step 7: The results obtained in Step 6... Perform vector decomposition to obtain the filter coefficient vector. Then, the filter coefficient vector The remote sensing image obtained after transformation domain filtering This enables remote sensing image detection of aquaculture cages.

[0062] Combination Figure 2 and Figure 3 A comparison of the two images revealed some local noise in the original image, and the seawater stratification showed obvious interference from ships. After processing with the algorithm of this invention for target detection of aquaculture cages, it was evident that the noise was suppressed to a certain extent, and the interference from ships was completely suppressed. In the original image, the aquaculture cages to be detected were densely distributed and had a similar spectrum to the seawater, exhibiting a "foreign object with the same spectrum" phenomenon. After processing with the algorithm of this invention, the aquaculture cages to be extracted were clearly extracted, the background noise was suppressed, no false positives or false negatives occurred, and the target contours were clearly extracted.

[0063] Example 2

[0064] This embodiment provides a remote sensing image aquaculture cage detection method based on an adaptive variable parameter dynamic learning network, including:

[0065] Obtain the original remote sensing image and construct the autocorrelation matrix of the original remote sensing image;

[0066] Specifically, a matrix representation of the original remote sensing image is obtained, and based on this matrix representation, the autocorrelation matrix of the original remote sensing image is obtained; wherein, the matrix representation is as follows:

[0067] ;

[0068] In the formula, It refers to the number of sensors. For a picture of size A grayscale image.

[0069] Based on the autocorrelation matrix, construct an equation-constrained optimization mathematical model;

[0070] Specifically, the mathematical model for the equality constraint optimization is as follows:

[0071] ;

[0072] in, These are the filter coefficients; It is the transpose operator; Let A be the target spectral vector and A be the autocorrelation matrix.

[0073] Based on the aforementioned equality constraint optimization mathematical model, a linear equality equation mathematical model is obtained;

[0074] Specifically, the constrained optimization mathematical model is transformed into an unconstrained optimization mathematical model using the Lagrange multiplier method, and a linear equation mathematical model is obtained based on the unconstrained optimization mathematical model; wherein, the linear equation mathematical model is as follows:

[0075] ;

[0076] In the formula, the autocorrelation coefficient matrix Coefficient vector ; Let be the vector to be solved; It is composed of filter coefficients. dimensional vector; It is a Lagrange multiplier.

[0077] Specifically, the unconstrained optimization model is as follows:

[0078] ;

[0079] in, It is a Lagrange multiplier.

[0080] An adaptive variable parameter dynamic learning network is used to solve the mathematical model of the linear equation, and the filtered remote sensing image is obtained based on the solution result to realize the detection of aquaculture cages in remote sensing images.

[0081] Specifically, the process of solving the mathematical model of the linear equation using an adaptive variable parameter dynamic learning network includes:

[0082] The parameters are initialized and an error function is constructed. The error function is calculated and iterated using adaptively changing parameters. It is determined whether the error function satisfies the loop condition. If the loop termination condition is met, the solution vector is output. If not, the adaptively changing parameters are updated and the error function is calculated again until the loop termination condition is met.

[0083] For example, by sampling the mathematical model of a linear equation, its corresponding discrete-time expression is obtained. Then, an adaptive variable parameter dynamic learning network is used to solve the mathematical model of the linear equation. The specific algorithm steps are as follows:

[0084] 1) Parameter initialization;

[0085] 1-1) Initialize allowable error ;

[0086] 1-2) Randomly generate initial points ;

[0087] 1-3) Initialize the step size factor Initialize the autocorrelation coefficient matrix. Initialize the coefficient vector Initialize the adaptive parameter expression. Initialize the number of iterations. ,make ;

[0088] 2) Calculate the error function ,like Stop calculation and output. ;

[0089] 3) Output based on adaptively changing parameters. ;in ,in This is the updated error function. It is achieved through adaptive parameter changes. Its function is to adaptively change parameters. Will follow The convergence of the time gradually amplifies and then shortens the exponential convergence time.

[0090] 4) Order . Physically, this can be understood as the direction of error adjustment, which is achieved through the autocorrelation matrix. Errors are weighted to control the step size and direction of parameter updates.

[0091] 5) Order , Proceed to step 2).

[0092] Specifically, the process of obtaining the filtered remote sensing image includes:

[0093] The vector to be solved is decomposed to obtain the filter coefficient vector; the filter coefficient vector is then transformed into the transform domain to obtain the filtered remote sensing image.

[0094] For example, the solution obtained Perform vector decomposition to obtain the filter coefficient vector. Then, the filter coefficient vector The remote sensing image obtained after transformation domain filtering This enables remote sensing image detection of aquaculture cages.

[0095] This embodiment also discloses a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0096] This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0097] This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0098] The adaptive variable parameter dynamic learning network proposed in this invention has the following distinct characteristics and advantages: 1) The network error function can take the form of a matrix, vector, or scalar depending on the problem type, and may be an unbounded real or complex number. By combining the error function with the evolution formula, the computational error of each element can converge to zero; 2) At the system level, the derivative information of the error function is utilized, causing the system to eventually evolve towards zero error and tend to stabilize; 3) An adaptive variable parameter is introduced into the design model. This adaptive variable parameter adjusts its changes adaptively as the error function converges, gradually amplifying and shortening the exponential convergence time, enabling rapid convergence and noise resistance. In summary, the adaptive variable parameter dynamic learning network proposed in this invention performs excellently in image classification and target detection, not only improving the accuracy of remote sensing image target detection but also shortening the computation time and enhancing target classification capabilities.

[0099] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting a fish cage in a remote sensing image based on a dynamic learning network with adaptive changing parameters, characterized in that, The method comprises the following steps: obtaining a raw remote sensing image and constructing an autocorrelation matrix of the raw remote sensing image; constructing an equation-constrained optimization mathematical model based on the autocorrelation matrix; obtaining a linear equation mathematical model based on the equation-constrained optimization mathematical model; solving the linear equation mathematical model by using an adaptive change parameter dynamic learning network, obtaining a filtered remote sensing image based on a solving result, and realizing remote sensing image aquaculture net cage detection; The process of solving the linear equation mathematical model by using the adaptive change parameter dynamic learning network comprises: performing parameter initialization and constructing an error function, calculating and iterating the error function by using adaptive change parameters, judging whether the error function meets a loop condition, outputting a to-be-solved vector if the loop end condition is met, and updating the adaptive change parameters and then calculating the error function again until the loop end condition is met if the loop end condition is not met; The solving process comprises the following steps: 1) parameter initialization; 1-1) Initialization of allowable error ; 1-2) Randomly generate initial points ; 1-3) initializing the step factor ; initializing the autocorrelation coefficient matrix ; initializing the coefficient vector ; initializing the adaptive variation parameter expression ; initializing the number of iterations , let ; 2) Calculate error function , if , stop calculation, output ; 3) output according to adaptive change parameter ; wherein , wherein are all adjustment factors, which can adjust the convergence speed of the model by adjustment; is the updated error function; through the role of adaptive change parameter , adaptive change parameter will gradually amplify with the convergence of and then shorten the exponential convergence time; 4) Let , The physical meaning of the adjustment direction of the error is that it weights the error by the autocorrelation matrix , thus controlling the step size and direction of the parameter update. 5) Let , , go to step 2).

2. The method of claim 1, wherein obtaining a matrix form representation of the raw remote sensing image, and obtaining an autocorrelation matrix of the raw remote sensing image based on the matrix form representation; wherein the matrix form representation is as follows: ； In the formula, is the number of sensors, is a gray scale image with a size of pixels.

3. The method of claim 1, wherein The equation-constrained optimization mathematical model is: ； wherein are filter coefficients; is a transpose operator; is a target spectral vector, A is an autocorrelation matrix.

4. The method of claim 3, wherein The equation-constrained optimization mathematical model is converted into an unconstrained optimization mathematical model by using a Lagrange multiplier method, and a linear equation mathematical model is obtained based on the unconstrained optimization mathematical model; wherein the linear equation mathematical model is as follows: ； where the autocorrelation coefficient matrix ; the coefficient vector ; is the vector to be solved; is a dimensional vector composed of filter coefficients; is a Lagrange function multiplier.

5. The method of claim 4, wherein The unconstrained optimization mathematical model is: ； wherein is a Lagrange multiplier.

6. The method of claim 1, wherein The process of obtaining the filtered remote sensing image comprises: performing vector decomposition on the to-be-solved vector to obtain a filter coefficient vector, and obtaining the filtered remote sensing image by transforming the filter coefficient vector in a transform domain.

7. A computer apparatus comprising: A memory and a processor for storing a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to realize the steps of the remote sensing image aquaculture net cage detection method based on the adaptive change parameter dynamic learning network according to any one of claims 1-6.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the remote sensing image aquaculture net cage detection method based on the adaptive change parameter dynamic learning network according to any one of claims 1-6.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the remote sensing image aquaculture net cage detection method based on the adaptive change parameter dynamic learning network according to any one of claims 1-6.

Images