Remote sensing image classification method based on membrane calculation, PSO particle swarm and BP neural network

Abstract

The invention belongs to the technical field of remote sensing data intelligent classification, and discloses a remote sensing image classification method based on membrane calculation, PSO particle swarm and BP neural network, which uses an organization type P system in membrane calculation as a calculation framework, and uses a speed-displacement model of the PSO particle swarm algorithm as an evolution rule of an object in a basic membrane. Meanwhile, sharing and co-evolution of information between objects are achieved through a transfer rule of a membrane, finally, a global optimal object is stored in the environment, then the global optimal object obtained in the environment serves as input of the optimized BP neural network, finally, the BP neural network is trained, and the optimal object is obtained. And inputting remote sensing image data into the trained BP neural network to obtain a final remote sensing image classification result. The remote sensing image classification efficiency and precision are improved, and rapid classification of the remote sensing images is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent classification of remote sensing data, and specifically relates to a remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network. Background Art

[0002] Membrane calculations were conducted by Gheorghe First proposed in 1998, the basic idea and principle of the algorithm is derived from a novel natural computing model that summarizes the structure and function of living cells and the mutual cooperation between cell groups in biological tissues and organs. This model can also be called P system or membrane system. Among them, the membrane is regarded as a type of protein molecule, including three states of ablation, division and production. Its permeability also changes at any time, and different changes correspond to different evolutionary rules, and different evolutionary rules correspond to different computing models. BP neural network has strong generalization ability. For learning the trained network and the extracted learning features, its application scope can not only predict the current existing samples, but also predict and judge the new sample data. However, the premise is to determine the complete structure and specific parameters of the BP neural network, and there must be a known label corresponding to the training samples to be learned.

[0003] At present, the existing remote sensing image classification methods are mainly applied to image classification software systems through deep learning and model training. This method is greatly affected by the system hardware environment resources, especially when processing large image files, which will seriously affect the speed of image classification. In addition, although some algorithms have fast convergence speed, they are prone to fall into local optimality. At the same time, image quality and image resolution will affect the accuracy of classification. In order to solve the shortcomings of the above technologies in processing large remote sensing image data, such as long time consumption, low accuracy, and falling into local optimality, the combination of "membrane computing + PSO particle swarm + BP neural network" algorithm is used to split the entire image classification task, search and optimize, and network training, which is particularly important for improving the efficiency and accuracy of image classification.

[0004] Existing technology "Remote sensing image classification based on quantum particle swarm algorithm for feature selection" proposes "a remote sensing image classification algorithm based on quantum particle swarm algorithm" to improve the classification effect of remote sensing images. By extracting various types of original features of remote sensing images, the quantum particle swarm algorithm is used to screen the features to extract the features that are more important to the classification results of remote sensing images; the remote sensing image classifier is established by using the least squares support vector machine (LSSVM) to achieve remote sensing image classification and recognition. Although the algorithm has high prediction accuracy and strong generalization ability, it requires a long training time and a large amount of calculation when processing a large amount of data, which to a certain extent limits its development in practical applications. "Research and Development of Peanut Hyperspectral Image Classification Method Based on SPA-PSO-BP" proposes a "classification detection model based on continuous projection algorithm (SPA) fused with particle swarm algorithm optimized BP neural network (PSO-BP)", using SPA to obtain characteristic wavelengths, and using PSO to optimize the initial weights and thresholds of the nonlinear modeling algorithm BP neural network to find the optimal solution, and construct a classification model based on the optimized BP modeling parameters combined with SPA. Although the algorithm can effectively reduce the model operation time, SPA causes a certain degree of original spectral data loss by reducing the redundant information between the wavelengths of the original spectral data and the characteristic wavelength of the minimum collinearity, which affects the accuracy of image classification. Summary of the invention

[0005] The purpose of the present invention is to disclose a remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network. Based on visible light remote sensing images, the PSO particle swarm optimization algorithm is integrated under the membrane computing framework to find the global optimal solution to optimize the classification detection model of the BP neural network, thereby realizing the rapid classification of remote sensing images and improving the efficiency and accuracy of remote sensing image classification.

[0006] The technical solution adopted by the present invention is a remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network, comprising the following steps:

[0007] S1, initialize the population parameters, the objects in each cell of the basic membrane, the number and speed of particles in the cell, and randomly distribute a set of equal number of initialized objects in each basic membrane;

[0008] S2, calculate the fitness value of each membrane object in the initial population according to the minimum mean square error criterion, and select the best object in all basic membranes and store it in the environment;

[0009] S3, iterative optimization training: the velocity-displacement model of the PSO particle swarm algorithm is used to evolve the objects in the basic membrane to obtain the evolved new objects; the local optimal objects in all basic membranes are compared and analyzed with the global optimal objects stored in the environment. If the local optimal object is better than the global optimal object stored in the environment, the global optimal object in the environment is updated through the transfer rule, otherwise the local optimal object is discarded;

[0010] S4, iterative optimization training termination judgment: if the maximum number of iterations of iterative optimization training is reached or the fitness function meets the convergence accuracy preset by the BP neural network, the iterative optimization training is terminated and the global optimal object in the environment is output; if the termination condition is not met, S2-S3 are repeated;

[0011] S5, directly assign the optimal weight and optimal threshold parameters corresponding to the global object obtained by iterative optimization in the above steps to the BP neural network as the initial weight and threshold of the BP neural network, then train the BP neural network, directly input the remote sensing image data into the BP neural network for simulation testing, and output the remote sensing image classification results predicted by the network simulation.

[0012] Furthermore, in S1, the P system used in the present invention is a tissue-type P system with q cells, which is expressed as:

[0013] ∏=(O,μ,R1,...,R q-1 ,R q ,R',i0)

[0014] Where q is the degree of the system ∏; O represents the set of objects in the cell, where each object represents a set of weight threshold parameters of the BP neural network to be optimized, that is, O = (a0, a1, ..., a L ,b1,b2,...,b M ), where a is the weight set of the BP neural network, b is the threshold set of the BP neural network, L is the maximum dimension of the BP neural network weight, M is the maximum dimension of the BP neural network threshold; μ is the membrane structure formed by q basic membranes; R i represents the set of evolutionary rules in the i-th cell, 1≤i≤q; R' represents the set of communication rules in q cells; 0 represents the environment, i0=0, indicating that environment 0 is the output area of ​​system ∏;

[0015] When initializing objects, it is pre-assumed that each of the q basic membranes contains a number of objects of the same number, and the dimension of each object is assumed to be D. The dimension of the object is expressed as D=L+M+1, where L is the maximum dimension of the BP neural network weight, and M is the maximum dimension of the BP neural network threshold. At the same time, Maxi and Mini are respectively assumed to represent the upper and lower bounds of the corresponding coefficients of the BP neural network parameters in the i-th dimension. Therefore, the initialization value of the parameter of the BP neural network corresponding to the i-th dimension, that is, the membrane object, is rand×(Maxi-Mini)+Mini, where rand is a random number between [0,1].

[0016] Furthermore, in S2, the present invention uses the minimum mean square error criterion to calculate and evaluate each object of each cell in the basic membrane, which is expressed as:

[0017]

[0018] Wherein, N is the number of remote sensing image samples; k represents the kth remote sensing image sample; d(k) is the predicted value output by the BP neural network of the kth remote sensing image sample, 1≤k≤N; y(k) is the actual value output by the BP neural network of the kth remote sensing image sample, 1≤k≤N;

[0019] The smaller the mean square error is, the better the parameters of the designed BP neural network are. Otherwise, the design effect is not good.

[0020] Furthermore, in S3, the present invention uses the velocity-displacement model of the PSO particle swarm algorithm as the evolution mechanism of the objects in each basic membrane, and uses the following two formulas to update the velocity and position of the objects in each basic membrane, thereby obtaining the global optimal object, that is, the weight and threshold parameters of the optimized BP neural network:

[0021]

[0022] Where i = 1, 2, ..., N, N is the number of samples; k is the number of iterations, k ≤ N; w represents the set of objects in multiple basic membranes in the designed organizational P system, that is, w is the parameter vector of the BP neural network, w = [a0a1a2...a L b1b2…b M ] T , where a is the weight set of the BP neural network, b is the threshold set of the BP neural network, L is the maximum dimension of the BP neural network weight, M is the maximum dimension of the BP neural network threshold, T is the dimension of the object in the basic membrane, T = L + M + 1; C1 and C2 are learning factors with values ​​between [0, 2], which are used to adjust the particle to its local optimal position P best Flight and global optimal position G gbestThe step length of flight; rand1∈(0,1), rand2∈(0,1) represent two random functions used to increase the randomness of particle flight.

[0023] Furthermore, in S3, there are two transfer rules of the present invention, wherein:

[0024] (1)Transshipment Rule 1:

[0025]

[0026] Where λ is an empty string, when j = i + 1, i = 1, 2, ..., q-1 or when j = 1, i = n, n is 0, the local optimal object of cell i itself is Communicate with its subsequent cell j through the ring membrane structure, so cell j will obtain its final optimal object, that is, the global optimal object Z gbest ;

[0027] (2)Transshipment Rule 2:

[0028]

[0029] Where λ is an empty string, i = 1, 2, ..., q, and the local optimal object of cell i itself is Transported to environment 0, and through this local optimal object Update the global optimal object Z in environment 0 gbest .

[0030] Furthermore, in the transport rule 2, the local optimal object of cell i itself is Update the global optimal object Z in environment 0 gbest , the update rule is:

[0031]

[0032] Among them, f(.) represents the mean square error MSE of an object; otherwise means if Then the global optimal object in environment 0 remains unchanged.

[0033] The beneficial effects of the present invention are:

[0034] (1) The present invention adopts an organizational P system in membrane computing as the multi-computing system framework for global optimization of the present invention, and utilizes the evolutionary rules and transport rules of the organizational P system to achieve information sharing and collaborative evolution between objects and between objects and the environment, thereby accelerating the global convergence speed of network parameters;

[0035] (2) The present invention can reduce the problem that the PSO particle swarm algorithm optimizes the BP neural network and easily falls into the local optimal solution;

[0036] (3) The present invention evolves the objects in the membrane by applying the velocity-displacement model of the PSO particle swarm algorithm as the evolution rule of the basic membrane, and uses the communication rules between cells to realize the communication and update between objects and between objects and the environment. It can effectively determine the optimal weights and threshold parameters of the BP neural network, improve the network training accuracy and shorten the network training time. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0038] Figure 2 Schematic diagram of the membrane structure of the tissue-type P system of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer and more understandable, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.

[0040] The present invention uses an organizational P system in membrane computing as a computing framework, and utilizes the velocity-displacement model of the PSO particle swarm algorithm as the evolution rule of objects in the basic membrane. At the same time, the transport rule of the membrane is used to realize the sharing and co-evolution of information between objects, and finally the global optimal object is saved in the environment, and then the global optimal object obtained in the environment is used as the input of the optimized BP neural network. Finally, the BP neural network is trained, and the remote sensing image data is input into the trained BP neural network to obtain the final remote sensing image classification result.

[0041] A remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network, such as Figure 1 As shown, the following steps are included:

[0042] S1, initialize the population parameters, the objects in each cell of the basic membrane, and the number and velocity of particles in the cell, and randomly distribute a set of equal number of initialized objects in each basic membrane.

[0043] The P system used in the present invention is a tissue P system (Tissue P System, TPS for short) with q cells (the degree of membrane structure q≥1), which is expressed as:

[0044] ∏=(O,μ,R1,...,R q-1 ,R q ,R',i0) (1)

[0045] Where O represents the set of objects in the cell, each of which represents a set of weight threshold parameters of the BP neural network to be optimized, that is, O = (a0, a1, ..., aL ,b1,b2,...,b M ), where a is the weight set of the BP neural network, b is the threshold set of the BP neural network, L is the maximum dimension of the BP neural network weight, and M is the maximum dimension of the BP neural network threshold; μ is a membrane structure formed by q basic membranes, and each membrane is identified by 1, 2, …, q in turn, where q is also called the degree of the system ∏; R i (1≤i≤q) represents the set of evolutionary rules in the i-th cell; R' represents the finite set of communication rules in the q-th cell, and each rule can be expressed as (i,u / v,j), where i is the i-th basic membrane, u represents the object u in the i-th basic membrane, j represents the j-th basic membrane, v represents the object v in the j-th basic membrane, i,j∈{0,1,...,q-1,q}, i≠j, 0 represents the environment; i0=0, indicating that the environment is the output area of ​​the system ∏.

[0046] like Figure 2 As shown, the membrane structure of the tissue-type P system of the present invention is a ring-shaped membrane structure, wherein the direction of the arrow represents the direction of the object transport between the basic membrane channels; at the same time, the object evolves in the cell through the evolution mechanism and adopts the transport rule to realize the communication and sharing of the object in the pre-specified channel. The environment is marked by 0 and serves as the output area of ​​the whole system.

[0047] Specifically, the object in each cell of the basic membrane is initialized by reading the input remote sensing image data, and the image is divided into a number of slices of the same size according to the principle of equal division.

[0048] In order to apply the organizational P system to solve the BP neural network optimization problem, the present invention uses the parameters of the BP neural network as objects in multiple basic membranes in the organizational P system, and its representation form is:

[0049] w=(a0,a1,...,a L ,b1,...,b M ) T (2)

[0050] Where T is the dimension of the BP neural network, T = L + M + 1, L is the maximum dimension of the BP neural network weight, and M is the maximum dimension of the BP neural network threshold; a i (0≤i≤L) is the weight of the BP neural network in the i-th cell; b j (1≤j≤M) is the threshold of the BP neural network in the jth cell.

[0051] When initializing objects, it is pre-assumed that each of the q basic membranes contains a number of objects of the same number, and that the dimension of each object is D. The dimension of the object is expressed as D=L+M+1, where L is the maximum dimension of the BP neural network weight, and M is the maximum dimension of the BP neural network threshold. At the same time, Maxi and Mini are respectively set to represent the upper and lower bounds of the corresponding coefficients of the BP neural network parameters in the i-th dimension. Therefore, the initialization value of the parameters of the BP neural network corresponding to the i-th dimension, that is, the membrane object, is rand×(Maxi-Mini)+Mini, where rand is a random number between [0,1].

[0052] S2, calculates the fitness value of each membrane object in the initial population according to the minimum mean square error criterion, and selects the best object from all basic membranes and stores it in the environment.

[0053] Specifically, assuming that each of the q basic membranes has an equal number of m initialization objects (remote sensing image slices), environment 0 is used as the final output area of ​​the system ∏ to store a global optimal object Z with the minimum mean square error (MSE) from the beginning to the end of the late iterative optimization training of the q basic membranes. gbest ; At the same time, the global optimal object Z is continuously updated in each iterative optimization training calculation process gbest , until the system ∏ shuts down, the final calculation result is saved to the global optimal object Z gbest middle.

[0054] Among them, the minimum mean square error criterion is used to calculate and evaluate each object of each cell in the basic membrane, which is expressed as:

[0055]

[0056] Among them, N is the number of remote sensing image samples, k is the kth remote sensing image sample, d(k)(0≤k≤N) is the predicted value output by the kth sample BP neural network; y(k)(0≤k≤N) is the actual value output by the kth sample BP neural network; the smaller the mean square error, the better the parameters of the designed BP neural network, otherwise, the design effect is not good.

[0057] S3, perform iterative optimization training: use the velocity-displacement model of the PSO particle swarm algorithm to evolve the objects in the basic membrane to obtain the evolved new objects; compare and analyze the local optimal objects in all basic membranes with the global optimal objects stored in the environment. If the local optimal object is better than the global optimal object stored in the environment, the global optimal object in the environment is updated through the transfer rule, otherwise the local optimal object is discarded.

[0058] The present invention uses the velocity-displacement model of the PSO particle swarm algorithm as the evolution mechanism of the objects in each basic membrane, and uses formulas (4) and (5) to update the velocity and position of the objects in each basic membrane (each object in the present invention corresponds to a particle in the PSO), thereby obtaining the global optimal object, that is, obtaining the weights and threshold parameters of the optimized BP neural network.

[0059]

[0060] Where i = 1, 2, ..., N, N is the number of samples; k is the number of iterations (k ≤ N); w represents the set of objects in multiple basic membranes in the designed organizational P system, that is, w is the parameter vector of the BP neural network, w = [a0a1a2...a L b1b2…b M ] T , where a is the weight set of the BP neural network, b is the threshold set of the BP neural network, L is the maximum dimension of the BP neural network weight, M is the maximum dimension of the BP neural network threshold, T is the dimension of the object in the basic membrane, T = L + M + 1; C1 and C2 are learning factors with values ​​between [0, 2], which are used to adjust the particle to its local optimal position P best Flight and global optimal position G gbest The step length of flight; rand1∈(0,1), rand2∈(0,1) represent two random functions used to increase the randomness of particle flight.

[0061] Specifically, assuming that there are N particles in the population, in a set D-dimensional problem search space, the position of the i-th (i=1, 2, ..., N) particle can be represented by a D-dimensional matrix X i =(x i1 ,x i2 ,...,x iD ) T To express it, the particle flight speed can be expressed as V i =(v i1 ,v i2 ,...,v iD ) T , and the fitness value of each particle in the population is calculated according to the objective function to evaluate the quality of the particle's position. The PSO particle swarm algorithm updates individuals by tracking two "extreme values", one of which is the optimal solution found by the individual itself, that is, the individual extreme value P best , denoted as P best =(p i1 ,p i2 ,...,p iD ), and the other is the optimal solution of the entire population so far, also called the global extreme value G gbest , denoted by G gbest=(p g1 ,p g2 ,...,p gD ).

[0062] The transport rule of the present invention mainly relies on the transport of objects between adjacent cells or between cells and the environment in the basic membrane to complete the exchange of objects and the sharing of information between different basic membranes and between the basic membrane and the environment, thereby obtaining the global optimal object Z gbest , and use the global optimal object Z in environment 0 gbest Update the worst object in each basic membrane in turn. Each cell communicates the local optimal object Z with its subsequent cells through the ring membrane structure. lbest , and at the same time, the local optimal object Z in the basic membrane lbest Then the global optimal object Z in environment 0 is updated through the transport rule of the membrane gbest .

[0063] The two transport rules used by the tissue-type P system designed by the present invention are as follows:

[0064] (1)Transshipment Rule 1:

[0065]

[0066] Where λ is an empty string, when j = i + 1, i = 1, 2, ..., q-1 or when j = 1, i = n, n is 0, the local optimal object of cell i itself is Communicate with its subsequent cell j through the ring membrane structure, so cell j will obtain its final optimal object, that is, the global optimal object Z gbest .

[0067] (2)Transshipment Rule 2:

[0068]

[0069] Where λ is an empty string, i = 1, 2, ..., q, and the local optimal object of cell i itself is Transported to environment 0, and through this local optimal object Update the global optimal object Z in environment 0 gbest , the update rule is shown in formula (8):

[0070]

[0071] Among them, f(.) represents the mean square error MSE of an object, otherwise means if Then the global optimal object in environment 0 remains unchanged.

[0072] S4, iterative optimization training termination judgment: if the maximum number of iterations of iterative optimization training is reached or the fitness function meets the convergence accuracy preset by the BP neural network, the iterative optimization training is terminated and the global optimal object in the environment is output; if the termination condition is not met, S2-S3 are repeated.

[0073] Among them, the global optimal object Z in the output environment gbest That is the optimal parameter required by the BP neural network of the present invention.

[0074] S5, directly assign the optimal weight and optimal threshold parameters corresponding to the global object obtained by iterative optimization in the above steps to the BP neural network as the initial weight and threshold of the BP neural network, then train the BP neural network, directly input the remote sensing image data into the BP neural network for simulation testing, and output the remote sensing image classification results predicted by the network simulation.

[0075] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network, characterized in that: The following steps are involved: S1, initialize the population parameters, the objects in each cell of the basic membrane, the number and speed of particles in the cell, and randomly distribute a set of equal number of initialized objects in each basic membrane; S2, calculate the fitness value of each membrane object in the initial population according to the minimum mean square error criterion, and select the best object in all basic membranes and store it in the environment; S3, iterative optimization training: the velocity-displacement model of the PSO particle swarm algorithm is used to evolve the objects in the basic membrane to obtain the evolved new objects; the local optimal objects in all basic membranes are compared and analyzed with the global optimal objects stored in the environment. If the local optimal object is better than the global optimal object stored in the environment, the global optimal object in the environment is updated through the transfer rule, otherwise the local optimal object is discarded; S4, iterative optimization training termination judgment: if the maximum number of iterations of iterative optimization training is reached or the fitness function meets the convergence accuracy preset by the BP neural network, the iterative optimization training is terminated and the global optimal object in the environment is output; If the termination condition is not met, repeat S2-S3; S5, directly assign the optimal weight and optimal threshold parameters corresponding to the global object obtained by iterative optimization in the above steps to the BP neural network as the initial weight and threshold of the BP neural network, then train the BP neural network, directly input the remote sensing image data into the BP neural network for simulation testing, and output the remote sensing image classification results predicted by the network simulation.

2. The remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network according to claim 1 is characterized in that: In S1, the P system used in the present invention is a tissue-type P system with q cells, which is expressed as: ∏(O,µ,R1,...,R q-1 ,R q ,R′,i0) Where q is the degree of the system ∏; O represents the set of objects in the cell; μ is the membrane structure formed by q basic membranes; R i represents the set of evolutionary rules in the i-th cell, 1≤i≤q; R' represents the set of communication rules in q cells; 0 represents the environment, i0=0, indicating that environment 0 is the output area of ​​system ∏; When initializing objects, it is pre-assumed that each of the q basic membranes contains a number of objects of the same number, and the dimension of each object is assumed to be D. The dimension of the object is expressed as D=L+M+1, where L is the maximum dimension of the BP neural network weight and M is the maximum dimension of the BP neural network threshold. At the same time, Maxi and Mini are respectively set to represent the upper and lower bounds of the corresponding coefficients of the BP neural network parameters in the i-th dimension. Therefore, the initialization value of the BP neural network parameter corresponding to the i-th dimension, that is, the membrane object, is rand×(Maxi-Mini+Mini, where rand is a random number between [0,1].

3. The remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network according to claim 1 is characterized in that: In S2, the present invention uses the minimum mean square error criterion to calculate and evaluate each object of each cell in the basic membrane, which is expressed as: Wherein, N is the number of remote sensing image samples; k represents the kth remote sensing image sample; d(k) is the predicted value output by the BP neural network of the kth remote sensing image sample, 1≤k≤N; y(k) is the actual value output by the BP neural network of the kth remote sensing image sample, 1≤k≤N; The smaller the mean square error is, the better the parameters of the designed BP neural network are. Otherwise, the design effect is not good.

4. The remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network according to claim 1 is characterized in that: In S3, the present invention uses the velocity-displacement model of the PSO particle swarm algorithm as the evolution mechanism of the objects in each basic membrane, and uses the following two formulas to update the velocity and position of the objects in each basic membrane, so as to obtain the global optimal object, that is, the weight and threshold parameters of the optimized BP neural network: Where i = 1, 2, ..., N, N is the number of samples; k is the number of iterations, k ≤ N; w represents the set of objects in multiple basic membranes in the designed organizational P system, that is, w is the parameter vector of the BP neural network, w = [a0a1a2...a L b1b2…b M ] T , where a is the weight set of the BP neural network, b is the threshold set of the BP neural network, L is the maximum dimension of the BP neural network weight, M is the maximum dimension of the BP neural network threshold, T is the dimension of the object in the basic membrane, T = L + M + 1; C1 and C2 are learning factors with values ​​between [0, 2], which are used to adjust the particle to its local optimal position P best Flight and global optimal position G gbest The step length of flight; rand1∈(0,1), rand2∈(0,1) represent two random functions used to increase the randomness of particle flight.

5. The remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network according to claim 1 is characterized in that: In S3, there are two types of transport rules of the present invention, wherein: (1)Transshipment Rule 1: Where λ is an empty string, when j = i + 1, i = 1, 2, ..., q-1 or when j = 1, i = n, n is 0, the local optimal object of cell i itself is Communicate with its subsequent cell j through the ring membrane structure, so cell j will obtain its final optimal object, that is, the global optimal object Z gbest ; (2)Transshipment Rule 2: Where λ is an empty string, i = 1, 2, ..., q, and the local optimal object of cell i itself is Transported to environment 0, and through this local optimal object Update the global optimal object Z in environment 0 gbest .

6. The remote sensing image classification method based on membrane computing + PSO particle swarm + BP neural network according to claim 5 is characterized in that , in the transport rule 2, through the local optimal object of cell i itself Update the global optimal object Z in environment 0 gbest , the update rule is: Among them, f(.) represents the mean square error MSE of an object; otherwise means if Then the global optimal object in environment 0 remains unchanged.

Images