Remote sensing image target detection method based on accelerated projection recurrent neural network

Abstract

The invention discloses a remote sensing image target detection method based on an accelerated projection recurrent neural network. The method comprises the following steps: acquiring an original remote sensing image and calculating a remote sensing image autocorrelation matrix; constructing a linear constraint optimization mathematical model of filtering output based on the remote sensing image autocorrelation matrix, and obtaining a linear equation mathematical model through conversion; and solving the linear equation mathematical model by adopting an accelerated projection recurrent neural network algorithm, carrying out vector decomposition on a solving result to obtain a filtering coefficient vector, obtaining a remote sensing image after filtering output based on the filtering coefficient vector, and completing remote sensing image target detection. According to the remote sensing image target detection method based on the accelerated projection recurrent neural network, the convergence speed is higher, the calculation time is shorter, the image threshold classification capability is higher and the target detection precision is higher when the optimization problem is solved.

Description

Technical Field

[0001] The invention belongs to the fields of remote sensing technology and image processing technology, and in particular relates to a remote sensing image target detection method based on an accelerated projection recursive neural network. Background Art

[0002] With the help of remote sensing detection, researchers and decision makers can accurately grasp the distribution and change trends of ground objects, which is of great significance to environmental monitoring, resource management, urban planning, agricultural development and other fields. Remote sensing technology has not only improved the ability to observe ground objects, but also promoted the rapid development of related research and application technologies. With the advancement of remote sensing technology, it is now possible to obtain images with high spectral and high spatial resolution, which has greatly promoted the research on ground object classification and extraction. How to effectively use remote sensing images to identify and monitor the spatiotemporal changes of surface objects has become one of the core research in this field. The need to grasp the distribution and change trends of ground objects has brought great challenges to the later remote sensing image processing technology, especially in the field of remote sensing image target detection. With the widespread application of remote sensing images, many classic target detection algorithms have been proposed in the past few decades and have been widely used. The core design of these algorithms is to reduce background interference, enhance target features, make the target more prominent in complex backgrounds, and facilitate identification. Representative algorithms include Match Filter (MF), Adaptive Coherence Estimator (ACE), and Constrained Energy Minimization (CEM). The complexity, information loss and uncertainty of the background of remote sensing images pose great challenges to the design of target detection algorithms. In practical applications, this background complexity often makes it difficult to extract accurate and reliable information. Therefore, most target extraction methods rely more on relatively easy-to-obtain and accurate prior target spectral information. However, ignoring the use of background spectral information may limit the improvement of detection accuracy and reliability. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a remote sensing image target detection method based on an accelerated projected recursive neural network to solve the problems existing in the above prior art.

[0004] To achieve the above object, the present invention provides a remote sensing image target detection method based on an accelerated projected recursive neural network, comprising:

[0005] Acquire the original remote sensing image and calculate the remote sensing image autocorrelation matrix; construct a linear constraint optimization mathematical model of the filter output based on the remote sensing image autocorrelation matrix; transform the linear constraint optimization mathematical model of the filter output to obtain a linear equation mathematical model;

[0006] The linear equation mathematical model is solved by using an accelerated projection recursive neural network algorithm, and the solution result is vectorized to obtain a filter coefficient vector. Based on the filter coefficient vector, a remote sensing image after filtering output is obtained to complete remote sensing image target detection.

[0007] Optionally, the process of calculating the autocorrelation matrix of the remote sensing image includes:

[0008] A matrix representation of the original remote sensing image is obtained, characteristic information of the original remote sensing image is obtained based on the matrix representation, and an autocorrelation matrix of the remote sensing image is calculated based on the characteristic information.

[0009] Optionally, the process of constructing a linear constrained optimization mathematical model of the filter output includes:

[0010] An optimal linear filter is constructed, wherein the optimal linear filter adopts a uniform gain constraint in the direction of the target spectral feature; a clutter power expression is constructed based on the remote sensing image autocorrelation matrix; and a linear constraint optimization mathematical model of the filter output is obtained based on the clutter power expression and the optimal linear filter.

[0011] Optionally, the linear constrained optimization mathematical model of the filter output is expressed as follows:

[0012]

[0013] In the formula, is the filter coefficient, is the target spectral vector, and Q is the autocorrelation matrix of the remote sensing image.

[0014] Optionally, the process of obtaining the linear equation mathematical model includes:

[0015] The Lagrange multiplier method is used to transform the linear constrained optimization mathematical model of the filter output into an unconstrained optimization mathematical model, and the unconstrained optimization mathematical model is transformed to obtain a linear equation mathematical model, which is expressed as:

[0016] ζφ(t)=κ;

[0017] Among them, the autocorrelation coefficient matrix Coefficient vector is the vector to be solved; w l =[w 1, w 2, w 3, …,w l ] T It is an l-dimensional vector composed of filter coefficients; λ(t) is the Lagrangian function multiplier.

[0018] Optionally, the process of solving the linear equation mathematical model using an accelerated projected recursive neural network algorithm includes:

[0019] The linear equation mathematical model is sampled to obtain a discrete-time expression of the mathematical model; parameters are initialized and an error function is constructed, and the discrete-time expression of the mathematical model is used as a basis, and the initialized parameters and the nonlinear accelerated projection function are combined to perform iterative solution, and whether a stopping condition is met is determined based on the calculation result of the error function. If not, the parameters are updated through the nonlinear accelerated projection function until the stopping condition is met.

[0020] Optionally, the parameter initialization process includes:

[0021] Initialize the allowable error and determine the stop condition; randomly generate an initial point as the starting point of the iteration, initialize the step factor, autocorrelation coefficient matrix, coefficient vector, nonlinear accelerated projection function expression and the number of iterations.

[0022] Optionally, the nonlinear accelerated projection function expression is:

[0023]

[0024] Among them, the design parameters v∈(0,1) and sign(·) represent the sign function, ∈ k Indicates error.

[0025] The present invention also provides a computer device, which includes a memory and a processor connected to the memory; the memory is used to store computer programs; the processor is used to run the computer programs stored in the memory to execute the steps of the above-mentioned remote sensing image target detection method based on accelerated projected recursive neural network.

[0026] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-mentioned remote sensing image target detection method based on accelerated projected recursive neural network.

[0027] Compared with the prior art, the present invention has the following advantages and technical effects:

[0028] The present invention provides a remote sensing image target detection method based on an accelerated projection recursive neural network. First, the autocorrelation matrix of the input original remote sensing image is calculated and a linear constraint optimization mathematical model of its filter output is established. Then, the linear equation mathematical model is converted by the Lagrange multiplier method, and the solution is obtained by using the accelerated projection recursive neural network algorithm. Finally, the output filter coefficient vector is passed through a change domain to obtain a remote sensing image after the filter output, thereby realizing remote sensing image target detection. The remote sensing image target detection method based on an accelerated projection recursive neural network proposed by the present invention has a faster convergence speed, a shorter calculation time, a stronger image threshold classification ability and a higher target detection accuracy when solving optimization problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] 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:

[0030] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0031] Figure 2 is an original remote sensing image according to an embodiment of the present invention;

[0032] Figure 3 This is an image after target detection processing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] 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.

[0034] 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.

[0035] Embodiment 1

[0036] like Figure 1 As shown, this embodiment provides a remote sensing image target detection method based on an accelerated projected recursive neural network, including:

[0037] Acquire the original remote sensing image and calculate the remote sensing image autocorrelation matrix; construct a linear constraint optimization mathematical model of the filter output based on the remote sensing image autocorrelation matrix; transform the linear constraint optimization mathematical model of the filter output to obtain a linear equation mathematical model;

[0038] In some embodiments, the process of calculating the remote sensing image autocorrelation matrix includes:

[0039] A matrix representation of the original remote sensing image is obtained, characteristic information of the original remote sensing image is obtained based on the matrix representation, and an autocorrelation matrix of the remote sensing image is calculated based on the characteristic information.

[0040] Specifically, for any remote sensing image, its matrix form can be expressed as where x i is the pixel vector in the remote sensing image, each pixel x i =[x i1, x i2, x i3, …,x im ] T It is an m-dimensional column vector, where m is the number of bands, where 1<<i<<N.

[0041] According to the characteristics of remote sensing images, the autocorrelation matrix can be calculated

[0042] In some embodiments, the process of constructing a linear constrained optimization mathematical model of the filtered output includes:

[0043] An optimal linear filter is constructed, wherein the optimal linear filter adopts a uniform gain constraint in the direction of the target spectral feature; a clutter power expression is constructed based on the remote sensing image autocorrelation matrix; and a linear constraint optimization mathematical model of the filter output is obtained based on the clutter power expression and the optimal linear filter.

[0044] In some embodiments, the linear constrained optimization mathematical model of the filtered output is expressed as follows:

[0045]

[0046] In the formula, is the filter coefficient, is the target spectral vector, and Q is the autocorrelation matrix of the remote sensing image.

[0047] Specifically, first establish the optimal linear filter w T x, and then under the uniform gain constraint in the target spectral feature direction, by minimizing the output clutter power w T Qw is established, and finally the linear constraint optimization mathematical model of the filter output is established;

[0048] In some embodiments, the process of obtaining the linear equation mathematical model includes:

[0049] The Lagrange multiplier method is used to transform the linear constrained optimization mathematical model of the filter output into an unconstrained optimization mathematical model, and the unconstrained optimization mathematical model is transformed to obtain a linear equation mathematical model, which is expressed as:

[0050] ζφ(t)=κ;

[0051] Among them, the autocorrelation coefficient matrix Coefficient vector is the vector to be solved; w l =[w 1, w 2, w 3, …,w l ] T It is an l-dimensional vector composed of filter coefficients; λ(t) is the Lagrangian function multiplier.

[0052] Specifically, the linear constraint optimization mathematical model of the filter output is transformed into an unconstrained optimization model through the Lagrange multiplier method, as shown below:

[0053] ψ(w)=w T Qw+λ(w T x-1);

[0054] Here, λ is the Lagrange multiplier.

[0055] The linear equation mathematical model is solved by using an accelerated projection recursive neural network algorithm, and the solution result is vectorized to obtain a filter coefficient vector. Based on the filter coefficient vector, a remote sensing image after filtering output is obtained to complete remote sensing image target detection.

[0056] In some embodiments, the process of solving the linear equation mathematical model using the accelerated projected recursive neural network algorithm includes:

[0057] The linear equation mathematical model is sampled to obtain a discrete-time expression of the mathematical model; parameters are initialized and an error function is constructed, and the discrete-time expression of the mathematical model is used as a basis, and the initialized parameters and the nonlinear accelerated projection function are combined to perform iterative solution, and whether a stopping condition is met is determined based on the calculation result of the error function. If not, the parameters are updated through the nonlinear accelerated projection function until the stopping condition is met.

[0058] In some embodiments, the process of parameter initialization includes:

[0059] Initialize the allowable error and determine the stop condition; randomly generate an initial point as the starting point of the iteration, initialize the step factor, autocorrelation coefficient matrix, coefficient vector, nonlinear accelerated projection function expression and the number of iterations.

[0060] Specifically, the linear equation mathematical model is sampled to obtain its corresponding discrete time expression. Then, the accelerated projected recursive neural network algorithm is used to solve it. The specific algorithm steps are as follows:

[0061] 1) Parameter initialization;

[0062] 1-1) Initialization tolerance ε = 10 -9 ;

[0063] 1-2) Randomly generate initial points

[0064] 1-3) Initialize the step size factor τ; initialize the autocorrelation coefficient matrix ζ; initialize the coefficient vector κ; initialize the nonlinear accelerated projection function expression Δ(·); initialize the number of iterations k, set k = 0;

[0065] 2) Calculate the error function If ‖< k ‖<ε, stop calculation and output

[0066] 3) According to the nonlinear accelerated projection function, output in The design parameters v∈(0,1) and sign(·) represent the sign function.

[0067] 4) Order

[0068] 5) Order k=k+1, go to step 2).

[0069] Specifically, the Perform vector decomposition to obtain the filter coefficient vector w k ; Then, the filter coefficient vector w k The remote sensing image after filtering output is obtained through the transform domain Realize target detection in remote sensing images.

[0070] This embodiment also provides a computer device, which includes a memory and a processor connected to the memory; the memory is used to store computer programs; the processor is used to run the computer programs stored in the memory to execute the steps of the above-mentioned remote sensing image target detection method based on accelerated projected recursive neural network.

[0071] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-mentioned remote sensing image target detection method based on accelerated projected recursive neural network.

[0072] As an important branch of the neural network, each neuron in the recursive neural network not only receives input from other neurons, but also outputs signals to other neurons, forming a dynamic system with feedback. The remote sensing image target detection method based on the accelerated projection recursive neural network proposed in the present invention has the following distinct characteristics and advantages: 1) Its error function can take a matrix, vector or scalar value according to the problem type, and is an indefinite and unbounded real number or complex number. Through the error function and the evolution formula, the calculation error of each element can converge to zero; 2) The derivative information of the time-varying coefficient is used at the system level, and it has a certain prediction ability; 3) It introduces a special nonlinear projection function in the design model, so that it can eliminate the time lag problem, achieve rapid convergence and have anti-noise ability. In summary, the method proposed in the present invention performs well in the field 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.

[0073] 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. After the target detection is performed by filtering with the algorithm of the present invention, it can be clearly seen that the noise has been suppressed to a certain extent. The objects (ships) that need to be detected on the original image are dense, the outline is blurred, and the image details are not very clear, and the resolution is low. After being processed by the method provided by the present invention, it can be clearly found that the target to be extracted is clearly extracted, and the background noise is suppressed, there is no missed detection or false detection, and the target outline is clearly extracted.

[0074] 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 remote sensing image target detection method based on accelerated projected recursive neural network, characterized in that: The following steps are involved: Obtain original remote sensing images and calculate the remote sensing image autocorrelation matrix; Constructing a linear constraint optimization mathematical model of the filter output based on the remote sensing image autocorrelation matrix; transforming the linear constraint optimization mathematical model of the filter output to obtain a linear equation mathematical model; The linear equation mathematical model is solved by using an accelerated projection recursive neural network algorithm, and the solution result is vectorized to obtain a filter coefficient vector. Based on the filter coefficient vector, a remote sensing image after filtering output is obtained to complete remote sensing image target detection.

2. The method for remote sensing image target detection based on accelerated projected recursive neural network according to claim 1, characterized in that: The process of calculating the autocorrelation matrix of remote sensing images includes: A matrix representation of the original remote sensing image is obtained, characteristic information of the original remote sensing image is obtained based on the matrix representation, and an autocorrelation matrix of the remote sensing image is calculated based on the characteristic information.

3. The method for remote sensing image target detection based on accelerated projected recursive neural network according to claim 1, characterized in that: The process of constructing a linear constrained optimization mathematical model of the filter output includes: An optimal linear filter is constructed, wherein the optimal linear filter adopts a uniform gain constraint in the direction of the target spectral feature; a clutter power expression is constructed based on the remote sensing image autocorrelation matrix; and a linear constraint optimization mathematical model of the filter output is obtained based on the clutter power expression and the optimal linear filter.

4. The method for remote sensing image target detection based on accelerated projected recursive neural network according to claim 3, characterized in that: The linear constrained optimization mathematical model of the filter output is expressed as follows: In the formula, is the filter coefficient, is the target spectral vector, and Q is the autocorrelation matrix of the remote sensing image.

5. The method for remote sensing image target detection based on accelerated projected recurrent neural network according to claim 4, characterized in that: The process of obtaining a mathematical model of a linear equation involves: The Lagrange multiplier method is used to transform the linear constrained optimization mathematical model of the filter output into an unconstrained optimization mathematical model, and the unconstrained optimization mathematical model is transformed to obtain a linear equation mathematical model, which is expressed as: ζφ(t)=κ; Among them, the autocorrelation coefficient matrix Coefficient vector is the vector to be solved; w l =[w 1, w 2, w 3, …,w l ] T It is an l-dimensional vector composed of filter coefficients; λ(t) is the Lagrangian function multiplier.

6. The method for remote sensing image target detection based on accelerated projected recurrent neural network according to claim 1, characterized in that: The process of solving the linear equation mathematical model using the accelerated projected recursive neural network algorithm includes: The linear equation mathematical model is sampled to obtain a discrete-time expression of the mathematical model; parameters are initialized and an error function is constructed, and the discrete-time expression of the mathematical model is used as a basis, and the initialized parameters and the nonlinear accelerated projection function are combined to perform iterative solution, and whether a stopping condition is met is determined based on the calculation result of the error function. If not, the parameters are updated through the nonlinear accelerated projection function until the stopping condition is met.

7. The method for remote sensing image target detection based on accelerated projected recurrent neural network according to claim 6, characterized in that: The parameter initialization process includes: Initialize the allowable error and determine the stop condition; randomly generate an initial point as the starting point of the iteration, initialize the step factor, autocorrelation coefficient matrix, coefficient vector, nonlinear accelerated projection function expression and the number of iterations.

8. The method for remote sensing image target detection based on accelerated projected recursive neural network according to claim 7, characterized in that: The nonlinear accelerated projection function expression is: Among them, the design parameters v∈(0,1) and sign(·) represent the sign function, ∈ k Indicates error.

9. A computer device, characterized in that: The computer device includes a memory and a processor connected to the memory; the memory is used to store a computer program; the processor is used to run the computer program stored in the memory to execute the steps of the remote sensing image target detection method based on an accelerated projected recursive neural network as described in any one of claims 1-8.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the remote sensing image target detection method based on an accelerated projected recursive neural network as described in any one of claims 1 to 8.

Images