Aquaculture net cage detection method based on adaptive variable parameter dynamic learning network

Abstract

The invention discloses a remote sensing image aquaculture net cage detection method based on a self-adaptive change parameter dynamic learning network, which comprises the following steps of: firstly, calculating an autocorrelation matrix of an input original remote sensing image and establishing an equality constraint optimization mathematical model of filtering output of the input original remote sensing image; then, a linear equation mathematical model is converted through a Lagrange multiplier method, a self-adaptive change parameter dynamic learning network algorithm is used for solving, finally, an output filter coefficient vector passes through a change domain to obtain a filtered and output remote sensing image, and remote sensing image aquaculture net cage detection is achieved. The adaptive variable parameter dynamic learning network provided by the invention has excellent performance in the fields of image classification and target detection, not only improves the precision of remote sensing image target detection, but also shortens the calculation time, and enhances the target classification capability.

Description

Technical Field

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

[0002] In recent years, the scale of domestic offshore aquaculture has continued to expand, offshore fishery aquaculture species are rich, and a variety of aquaculture methods based on cages are adopted. Technological progress has significantly improved aquaculture efficiency and product quality. However, since cage aquaculture methods are mostly distributed in shallow areas with a water depth of less than 20 meters, the terrain is relatively closed and the exchange capacity with offshore waters is poor. Feed residues, fish feces and drug residues in the aquaculture process will lead to environmental problems such as eutrophication and algal blooms. Furthermore, some sea areas have too dense aquaculture cages and lack scientific planning, which further aggravates the environmental problems of the water. Therefore, it is necessary to conduct precise inspections of the aquaculture cage layout and surrounding waters. The traditional method of relying on manual inspections is difficult to meet the monitoring needs of modern large-scale precision aquaculture due to insufficient coverage and efficiency. Therefore, the introduction of remote sensing technology has become an inevitable trend.

[0003] Remote sensing obtains environmental information through satellites or aircraft. It has the advantages of wide coverage, high real-time performance and low cost. It can be used for efficient monitoring and management of aquaculture areas and promote the sustainable development of aquaculture. Remote sensing technology overcomes many defects of traditional monitoring methods and has higher spatial freedom and development potential. With the continuous updating of satellite remote sensing sensors, high-spectral and high-spatial resolution remote sensing image technology has developed rapidly, making the classification and extraction of ground objects a research focus. From early visual interpretation to classical classification methods, to neural networks and deep learning algorithms, remote sensing aquaculture area extraction has achieved remarkable results. Effective extraction usually relies on the recognition and segmentation of remote sensing image data and relies on intelligent algorithms, including pixel classification, object-oriented methods and neural networks. Due to the complex background in remote sensing images and the lack and uncertainty of information, the design of target detection algorithms has brought great challenges. This complex background often makes it difficult to obtain accurate and reliable information. Therefore, most target extraction methods rely on easy-to-obtain and high-precision prior target spectral information. However, ignoring the use of background spectral information may limit the further improvement of detection accuracy and reliability. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a remote sensing image aquaculture cage detection method based on a dynamic learning network with adaptive variable parameters to solve the problems existing in the above prior art.

[0005] To achieve the above object, the present invention provides a method for detecting aquaculture cages using remote sensing images based on a dynamic learning network with adaptively changing parameters, comprising:

[0006] Obtaining the original remote sensing image and constructing the autocorrelation matrix of the original remote sensing image;

[0007] Constructing an equality-constrained optimization mathematical model based on the autocorrelation matrix;

[0008] Obtaining a linear equality equation mathematical model based on the equality constraint optimization mathematical model;

[0009] The linear equation mathematical model is solved by using a dynamic learning network with adaptively changing parameters, and a remote sensing image after filtering output is obtained based on the solution result, so as to realize the detection of aquaculture cages using remote sensing images.

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

[0011] ;

[0012] In the formula, is the number of sensors, For a size of Grayscale image of .

[0013] Optionally, the equality constraint optimization mathematical model is:

[0014] ;

[0015] in, is the filter coefficient; is the transposition operator; is the target spectrum vector, and A is the autocorrelation matrix.

[0016] Optionally, the equation-constrained optimization mathematical model is converted into an unconstrained optimization mathematical model 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:

[0017] ;

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

[0019] Optionally, the unconstrained optimization model is:

[0020]

[0021] in, is the Lagrange multiplier.

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

[0023] Initialize the parameters and construct the error function, use the adaptive change parameters to iterate the error function, and determine whether the error function meets the loop condition. If the loop end condition is met, output the vector to be solved; if not, update the adaptive change parameters and calculate the error function again until the loop end condition is met.

[0024] Optionally, the process of obtaining the remote sensing image after filtering output includes:

[0025] The vector to be solved is decomposed to obtain a filter coefficient vector; the filter coefficient vector is transformed into a remote sensing image after filtering output.

[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 on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

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

[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 the form of a matrix, vector or scalar depending on the problem type, and may be an indefinite and unbounded real or complex number. By combining the error function with the evolution formula, the computational error of each element can be converged to zero;

[0031] 2) The derivative information of the error function is used at the system level so that the system will eventually evolve in the direction of zero error and tend to be stable;

[0032] 3) An adaptive variable parameter is introduced in the design model. The adaptive variable parameter will adaptively adjust the change of the parameter as the error function converges, gradually amplifying and then shortening the convergence time exponentially, so that it can achieve fast convergence and have noise resistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0034] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0035] Figure 2 is an original remote sensing image according to an embodiment of the present invention;

[0036] Figure 3 This is an image after the aquaculture cage target detection processing is performed by this method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

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

[0039] Embodiment 1

[0040] like Figure 1-3 As shown, in this embodiment, a method for detecting aquaculture cages using remote sensing images based on a dynamic learning network with adaptively changing parameters is provided, comprising:

[0041] Step 1: For any remote sensing image, its matrix form can be expressed as ,in is the number of sensors, For a size of Grayscale image of .

[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] is the filter coefficient; is the transposition operator; is the target spectrum vector.

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

[0047] (2)

[0048] in, is the Lagrange multiplier.

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

[0050] (3)

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

[0052] Step 6: Sampling the linear equation mathematical model (3) of step 5 to obtain its corresponding discrete time expression. Then, using the adaptive parameter dynamic learning network, the linear equation mathematical model (3) of step 5 is solved. The specific algorithm steps are as follows:

[0053] 6-1) Parameter initialization;

[0054] 6-1-1) Initialization tolerance ;

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

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

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

[0058] 6-3) According to the adaptive change parameters, output ;in ,in is the updated error function. By adaptively changing the parameters The role of adaptive change parameters Will follow The convergence of the function is gradually amplified and then the convergence time is shortened exponentially.

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

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

[0061] Step 7: Perform vector decomposition to obtain the filter coefficient vector ; Then, the filter coefficient vector The remote sensing image after filtering output is obtained through the transform domain , realizing remote sensing image aquaculture cage detection.

[0062] Combination Figure 2 and Figure 3 By comparing the two images as a whole, it is found that the original image contains some local noise, and the seawater stratification is obviously disturbed by ships. After the aquaculture cage target detection is performed by the algorithm of the present invention, it can be clearly seen that the noise has been suppressed to a certain extent, and the interfering ships have been completely suppressed. The aquaculture cages that need to be detected on the original image are densely distributed and similar to the seawater spectrum, and there is a "different objects with the same spectrum" phenomenon. After being processed by the algorithm of the present invention, it can be clearly found that the aquaculture cages to be extracted are clearly extracted, and the background noise is suppressed, there is no missed detection or false detection, and the target contour is clearly extracted.

[0063] Embodiment 2

[0064] This embodiment provides a method for detecting aquaculture cages using remote sensing images based on a dynamic learning network with adaptively changing parameters, including:

[0065] Obtaining the original remote sensing image and constructing 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 the matrix representation, an autocorrelation matrix of the original remote sensing image is obtained; wherein the matrix representation is as follows:

[0067] ;

[0068] In the formula, is the number of sensors, For a size of Grayscale image of .

[0069] Constructing an equality-constrained optimization mathematical model based on the autocorrelation matrix;

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

[0071] ;

[0072] in, is the filter coefficient; is the transposition operator; is the target spectrum vector, and A is the autocorrelation matrix.

[0073] Obtaining a linear equality equation mathematical model based on the equality constraint optimization mathematical model;

[0074] Specifically, the equation constraint optimization mathematical model is converted 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 ; is the vector to be solved; It is composed of filter coefficients dimensional vector; is the Lagrange function multiplier.

[0077] Specifically, the unconstrained optimization model is:

[0078] ;

[0079] in, is the Lagrange multiplier.

[0080] The linear equation mathematical model is solved by using a dynamic learning network with adaptively changing parameters, and a remote sensing image after filtering output is obtained based on the solution result, so as to realize the detection of aquaculture cages using remote sensing images.

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

[0082] Initialize the parameters and construct the error function, use the adaptive change parameters to iterate the error function, and determine whether the error function meets the loop condition. If the loop end condition is met, output the vector to be solved; if not, update the adaptive change parameters and calculate the error function again until the loop end condition is met.

[0083] Exemplarily, by sampling the linear equation mathematical model, its corresponding discrete time expression is obtained. Then, the linear equation mathematical model is solved by using the adaptive parameter dynamic learning network. The specific algorithm steps are as follows:

[0084] 1) Parameter initialization;

[0085] 1-1) Initialization tolerance ;

[0086] 1-2) Randomly generate the initial point ;

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

[0088] 2) Calculate the error function ,like , stop calculation, output ;

[0089] 3) According to the adaptive change parameters, output ;in ,in is the updated error function. By adaptively changing the parameters The role of adaptive change parameters Will follow The convergence of the function is gradually amplified and then the convergence time is shortened exponentially.

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

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

[0092] Specifically, the process of obtaining the remote sensing image after filtering output includes:

[0093] The vector to be solved is decomposed to obtain a filter coefficient vector; the filter coefficient vector is transformed into a remote sensing image after filtering output.

[0094] For example, the solution Perform vector decomposition to obtain the filter coefficient vector ; Then, the filter coefficient vector The remote sensing image after filtering output is obtained through the transform domain , realizing remote sensing image aquaculture cage detection.

[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, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

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

[0098] The adaptive variable parameter dynamic learning network proposed in the present invention has the following distinct characteristics and advantages: 1) The network error function can take the form of a matrix, vector or scalar according to the problem type, and may be an indefinite and unbounded real number or complex number. By combining the error function with the evolution formula, the calculation error of each element can be converged to zero; 2) The derivative information of the error function is used at the system level, so that the system will eventually evolve in the direction of zero error and tend to be stable; 3) It introduces an adaptive variable parameter in the design model. The adaptive variable parameter will gradually amplify and shorten the exponential convergence time as the error function converges and the parameter changes adaptively, so that it can achieve rapid convergence and have noise resistance. In summary, the adaptive variable parameter dynamic learning network proposed in the present invention performs well in the fields of image classification and target detection, not only improving the accuracy of remote sensing image target detection, but also shortening the calculation time and enhancing the target classification ability.

[0099] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for detecting aquaculture cages from remote sensing images based on a dynamic learning network with adaptively changing parameters, characterized in that: The following steps are involved: Obtaining the original remote sensing image and constructing the autocorrelation matrix of the original remote sensing image; Constructing an equality-constrained optimization mathematical model based on the autocorrelation matrix; Obtaining a linear equality equation mathematical model based on the equality constraint optimization mathematical model; The linear equation mathematical model is solved by using a dynamic learning network with adaptively changing parameters, and a remote sensing image after filtering output is obtained based on the solution result, so as to realize the detection of aquaculture cages using remote sensing images.

2. The method according to claim 1, characterized in that Obtain a matrix representation of the original remote sensing image, and based on the matrix representation, obtain an autocorrelation matrix of the original remote sensing image; wherein the matrix representation is as follows: ； In the formula, is the number of sensors, For a size of Grayscale image of .

3. The method according to claim 1, characterized in that The equality constraint optimization mathematical model is: ； in, is the filter coefficient; is the transposition operator; is the target spectrum vector, and A is the autocorrelation matrix.

4. The method according to claim 3, characterized in that The equation-constrained optimization mathematical model is converted 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: ; In the formula, the autocorrelation coefficient matrix ; Coefficient vector ; is the vector to be solved; It is composed of filter coefficients dimensional vector; is the Lagrange function multiplier.

5. The method according to claim 4, characterized in that The unconstrained optimization model is: in, is the Lagrange multiplier.

6. The method according to claim 1, characterized in that The process of solving the linear equation mathematical model using an adaptively variable parameter dynamic learning network includes: Initialize the parameters and construct the error function, use the adaptive change parameters to iterate the error function, and determine whether the error function meets the loop condition. If the loop end condition is met, output the vector to be solved; if not, update the adaptive change parameters and calculate the error function again until the loop end condition is met.

7. The method according to claim 6, characterized in that The process of obtaining the remote sensing image after filtering output includes: The vector to be solved is decomposed to obtain a filter coefficient vector; the filter coefficient vector is transformed into a remote sensing image after filtering output.

8. 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 remote sensing image aquaculture cage detection method based on an adaptively changing parameter dynamic learning network as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the remote sensing image aquaculture cage detection method based on an adaptive variable parameter dynamic learning network described in any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the remote sensing image aquaculture cage detection method based on an adaptive variable parameter dynamic learning network described in any one of claims 1 to 7 are implemented.

Images