1. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm is characterized by comprising the following steps of:

step 1, acquiring two forest remote sensing images I at different moments₁And I₂Respectively preprocessing each forest remote sensing image to obtain two preprocessed forest remote sensing images I₁"and I₂"; wherein, I₁And I₂The sizes of the images are the same, and are M multiplied by N multiplied by B, M, N respectively represent the number of rows and columns of the images, B represents the number of spectral bands of the images, M is more than or equal to 20, N is more than or equal to 20, and B is more than or equal to 4;

step 2, respectively calculating two preprocessed forest remote sensing images I by utilizing the reflectivity of the spectral band₁″、I₂"of normalized vegetation index matrix NDVI₁、NDVI₂And making it be the corresponding vegetation index characteristic diagram SF₁、SF₂；

Step 3, vegetation index characteristic diagram SF₁And SF₂Respectively carrying out median filtering to correspondingly obtain denoised feature maps SF₁′、SF₂'; denoising feature map SF by using difference method₁′、SF₂' construct vegetation index disparity map DI;

step 4, carrying out significance detection on the vegetation index difference map DI to obtain a significance difference map DS;

step 5, pre-classifying the significant difference image DS by using a fuzzy C-means clustering algorithm, dividing pixels in the DS into a variation class, an unchanged class and an uncertain class, and correspondingly obtaining a variation class pixel set, an unchanged class pixel set and an uncertain class pixel set;

step 6, optimizing the class labels of the pixels in the uncertain class pixel set by using a genetic algorithm based on self-adaptive parameters to obtain optimized class labels corresponding to the uncertain classes; and generating a final change graph according to the optimized class labels.

2. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm as claimed in claim 1, wherein in step 1, the preprocessing is pair I₁And I₂Respectively carrying out radiation correction, and then carrying out atmospheric correction on the images after the radiation correction.

3. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm according to claim 2, wherein in the step 2, a calculation formula of the normalized vegetation index matrix is as follows:

where ρ is_NIRAnd ρ_REDRespectively representing the infrared spectrum band reflectivity and the red spectrum band reflectivity of the preprocessed forest remote sensing image;

enabling each preprocessed forest remote sensing image I₁″、I₂"derived normalized vegetation index matrix NDVI₁、NDVI₂Vegetation index profile SF for corresponding forest image₁、SF₂The image size is M × N.

4. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm as claimed in claim 1, wherein the denoised feature map SF is subjected to a difference method₁′、SF₂' the constructed vegetation index difference map DI is specifically: de-noised characteristic diagram SF₁′、SF₂' the pixel values of the corresponding positions are subtracted, and the difference result is used as the pixel values of the corresponding positions of the exponential difference map DI.

5. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm according to claim 1, characterized in that the significance detection of the vegetation index difference map DI is carried out by the specific process:

4.1, calculating the global contrast of each pixel (x, y) in the vegetation index difference map DI on DI, and taking the sum of euclidean distances of the pixel and all other pixels (i, j) in the DI on the color as the saliency characteristic of the pixel, wherein the calculation formula is as follows:

wherein Sam (x, y) is a saliency feature of pixel (x, y), and (x, y) ≠ (i, j); | l | · | | represents solving the euclidean distance; then normalizing Sam (x, y) to [0,255] to obtain a significance map Sam;

4.2, generating a binaryzation significance difference map S by using a maximum inter-class variance method for the significance map Sam; and then generating a significance difference map DS by using the binarized significance map S and a vegetation index difference map DI:

DS(x，y)＝S(x，y)·DI(x，y)。

6. the forest change remote sensing detection method based on the adaptive parameter genetic algorithm according to claim 1, wherein the significance difference map DS is pre-classified by using a fuzzy C-means clustering algorithm, and specifically comprises the following steps:

5.1, setting an objective function of a fuzzy C-means clustering algorithm as follows:

wherein c is the number of categories, n is the total number of pixels of the significant difference map DS, u_ijRepresentative pixel x_jMembership degree belonging to the ith cluster, and the value is [0,1 ]]And is andm represents a blurring weight coefficient, d_ij＝||x_j-v_i| l, representing pixel x_jAnd the clustering center v_iThe Euclidean distance of;

c, m, a threshold epsilon and a maximum iteration number T are set; initializing the iteration number t to 1, and randomly initializing three cluster centers v₁、v₂、v₃；

5.2, calculating a new clustering center and membership function:

if the current iteration algebra t is 1, directly updating the membership degree of each pixel according to the pixel value of the clustering center; if the current iteration algebra t is more than or equal to 2, updating the clustering center by using the membership matrix obtained by the previous generation, and updating the membership matrix according to the clustering center; the concrete formula is as follows:

wherein k is 1,2, 3;representing the ith cluster center in the t iteration,representing pixel x in the t-th iteration_jMembership belonging to the ith cluster;

5.3, ifAnd T is less than or equal to T, making T equal to T +1, and returning to the step 5.2; otherwise, iteration is terminated, and the class label is distributed according to the final membership degree of each pixel: for pixel x_jIf arg_k{max{u_kj}}＝arg_k{max{v_k} and max { u }_kjIf the pixel is equal to or larger than 0.9, the pixel belongs to the variation class; arg_k{max{v_kDenotes the class k, arg with the largest cluster center pixel value_k{max{u_kjDenotes pixel x_jThe category corresponding to the maximum membership degree of (c); if arg_k{max{u_kj}}＝arg_k{min{v_k} and max { u }_kjIf the pixel is not less than 0.9, the pixel belongs to the unchanged class; otherwise, the pixel belongs to the uncertain class.

7. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm according to claim 1, wherein the adaptive parameter genetic algorithm is used for optimizing the class labels of the pixels in the uncertain class pixel set, and specifically comprises the following steps:

6.1, population initialization: setting the size of a population to be Num, namely, Num individuals coded into a binary two-dimensional matrix are in the population, and the size of an individual matrix Ind is the same as that of a significance difference map DS; initializing all individual matrixes Ind in the population according to the class labels obtained in the step 5, namely, modifying the pixel value of the position corresponding to the changed pixel set to 1, modifying the pixel value of the position corresponding to the unchanged pixel set to 0, and randomly coding the pixel value of the position corresponding to the uncertain pixel set to 0 or 1; establishing a position index of the uncertain pixels in the individual matrix, and setting the iteration number K 'as 1, the threshold epsilon' and the maximum iteration number K;

6.2, calculating individual fitness:

introducing spatial information OF pixels, designing an objective function OF with a neighborhood factor:

wherein, r is a category identifier, r is 0,1, r is 0 corresponding to the unchanged category, and r is 1 corresponding to the changed category; c. C_rAll pixels representing the corresponding class in the salient region, n_rAnd m_rRepresenting the statistical number and average gray value, N, of pixels of the corresponding category_qRepresenting the central pixel x_j3 x 3 neighborhood N_jOf (1) the q-th pixel, l_qIs N_qAnd a central pixel x_jIs the Euclidean distance of L is the pixel x_jDistance weights to its surrounding pixels; DS (direct sequence)_jRepresenting the center pixel x in the significant difference map DS_jPixels corresponding to the positions;

after the objective function of each individual is obtained, the Fitness Fitness of each individual is obtained by taking the reciprocal of the objective function:

6.3, performing elite selection and cross operation on the population:

if the current iteration algebra k' is 1, performing genetic operation on all individuals; if the current iteration algebra k 'is more than or equal to 2, the current iteration individual is an offspring individual, the k' -1 generation individual is a parent individual, Num individuals are selected from high to low according to the fitness values of all the individuals of the offspring and the parent to execute genetic operation, the individual with the highest fitness value is selected as the best individual of the current iteration algebra, the best fitness value is stored in a matrix Maxfit, and the unselected individuals are eliminated in evolution; then, performing cross operation on the selected individuals performing the genetic operation to obtain a crossed population;

6.4, calculating the variation probability of each individual in the crossed population pixel by pixel:

if the obtained mutation probability p (x)_j) If the value is greater than the preset standard variation probability p, the code of the position is changed, namely 0 is changed into 1 or 1 is changed into 0, otherwise, the code value is not changed; traversing each individual matrix to obtain a varied population, and taking the varied population as a progeny population of the previous generation; calculating the fitness of each individual in the offspring population, and selecting the fitness with the maximum fitness as the optimal fitness of the offspring;

6.5, if the difference of the best fitness obtained by two adjacent iteration algebras is larger than a set threshold value, namely | Maxfit_k′+1-Maxfit_k′If | ≧ epsilon 'and K' is less than or equal to K, making K '═ K' +1, returning to step 6.3, otherwise, outputting the individual with the highest fitness of the iteration; and determining the category label of the optimized uncertain pixel set according to the individual.

8. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm according to claim 7, wherein the cross operation is performed on the individuals selected to perform the genetic operation, and the specific method comprises the following steps: firstly, grouping all individuals executing genetic operation pairwise to form a plurality of groups of individuals to be crossed; then randomly at (0,1)Initializing a matrix A with the same size as the individual, traversing the elements of the matrix A, and when the value of a certain element is less than the preset cross probability p_cMarking the position of the element; and then exchanging elements in each group of individuals to be crossed, which are the same as the marked positions, and keeping other positions unchanged to form a crossed population.

9. The forest change remote sensing detection method based on the adaptive parameter genetic algorithm according to claim 7, wherein the variation probability of each individual in the crossed population is calculated pixel by pixel, and the specific process is as follows:

for each individual in the crossed population, traversing the individual matrix and aiming at a target pixel x_j3 x 3 neighborhood L_jEach neighborhood pixel L in_j(s), s 1,2,3, 9, whose membership u belongs to the class r is calculated_sr：

Wherein r is 0, 1: m is_rRepresenting the average gray value of each type of pixel obtained in the step 6.2;

according to the degree of membership u_srIs a neighborhood pixel L_j(s) assigning a class label if u_s0≥u_s1Then c is_sNot equal to 0, otherwise, c_s1 is ═ 1; thus, the target pixel x_jThe mutation probability of (c) is:

wherein d is_sIs a neighborhood pixel L_j(s) to center pixel point x_jSpatial distance of (B)_jIs x_jThe binary coded value of (a).

Background

Change detection is one of the most important research subjects in remote sensing image processing, means that two or more remote sensing images are acquired at different times in the same geographical area, qualitative and quantitative analysis and measurement of surface changes are carried out, and the method is widely applied to the fields of environment monitoring, natural disaster assessment, forest resource monitoring, agricultural investigation and the like.

Forest covers 31% of the earth's surface and is the largest ecosystem on human land. The forest has important influence on ecological environment, biodiversity and climate change, plays an important role in the aspects of purifying air, adjusting climate, conserving water source, reducing wind and sand harm and the like, and the quantity and quality of the forest become important material bases of national economy. The change of the forest resource condition has important influence on the development progress of the country, even the global ecological environment, the biological diversity and the climate change. Therefore, the forest coverage change information is timely and accurately acquired, and the method has important significance for researching environment change and forest management planning.

The traditional forest change detection method mainly adopts manual visual interpretation, needs special interpretation personnel to compare and interpret the change areas of images in different periods, and is difficult to quickly and accurately distinguish the forest areas in a large range. Compared with the traditional method, the remote sensing technology has the advantages of large detection range, short data acquisition period, less limitation by ground conditions and the like, and is widely applied to the field of forest change detection in recent years. Remote sensing images of different plants in different seasons are greatly different, and seasonal differences of vegetation can be overcome by utilizing spectral characteristic parameters such as normalized vegetation indexes NDVI, ratio vegetation indexes RVI and the like to detect forest changes. However, since the remote sensing image is subjected to noise interference of different degrees in the collection and transmission processes, the classification accuracy is low only by using the spectral features to classify the remote sensing image.

Disclosure of Invention

Aiming at the problems in the prior art, the invention aims to provide a forest change remote sensing detection method based on a self-adaptive parameter genetic algorithm, which is used for solving the problems of low forest change detection precision and huge search space caused by the genetic algorithm using a single spectral feature.

In order to achieve the purpose, the invention is realized by adopting the following technical scheme.

The forest change remote sensing detection method based on the adaptive parameter genetic algorithm comprises the following steps:

step 1, acquiring two forest remote sensing images I at different moments₁And I₂Respectively preprocessing each forest remote sensing image to obtain two preprocessed forest remote sensing images I₁"and I₂”；

Step 2, respectively calculating two preprocessed forest remote sensing images I by utilizing the reflectivity of the spectral band₁”、I₂"normalized vegetation index matrix NDVI₁、NDVI₂And making it be the corresponding vegetation index characteristic diagram SF₁、SF₂；

Step 3, vegetation index characteristic diagram SF₁And SF₂Respectively carrying out median filtering to correspondingly obtain denoised feature maps SF₁'、SF₂'; denoising feature map SF by using difference method₁'、SF₂' construct vegetation index disparity map DI;

step 4, carrying out significance detection on the vegetation index difference map DI to obtain a significance difference map DS;

step 5, pre-classifying the significant difference image DS by using a fuzzy C-means clustering algorithm, dividing pixels in the DS into a variation class, an unchanged class and an uncertain class, and correspondingly obtaining a variation class pixel set, an unchanged class pixel set and an uncertain class pixel set;

step 6, optimizing the class labels of the pixels in the uncertain class pixel set by using a genetic algorithm based on self-adaptive parameters to obtain optimized class labels corresponding to the uncertain classes; and generating a final change graph according to the optimized class labels.

Compared with the prior art, the invention has the beneficial effects that:

(1) the invention uses significance detection to the filtered vegetation index difference map DI to obtain the approximate position of the change region, and uses fuzzy C-means clustering algorithm to pre-classify in the significance region, thereby greatly reducing the search space of evolutionary computation;

(2) the method selects a genetic algorithm to optimize the pixel label of the forest remote sensing image, carries out multi-point search in the whole situation, has large coverage area and is not easy to fall into local optimum;

(3) the genetic algorithm used by the invention not only considers the spectral information of the image, but also introduces a spatial neighborhood factor, thereby solving the problem of poor noise immunity of the traditional genetic algorithm on the forest change detection task; meanwhile, the genetic operator with self-adaptive change enables the mutation probability of the pixel to be changed in a self-adaptive mode, and the convergence speed of the algorithm is accelerated.

Drawings

The invention is described in further detail below with reference to the figures and specific embodiments.

FIG. 1 is a flow chart of an implementation of the present invention;

FIG. 2 is a processing result of a forest remote sensing image and different algorithms of the simulation experiment of the present invention; wherein, (a) and (b) are forest remote sensing images at different moments respectively; (c) the variation graph obtained by the original genetic algorithm, and (d) the variation graph obtained by the method of the invention.

Detailed Description

Embodiments of the present invention will be described in detail below with reference to examples, but it will be understood by those skilled in the art that the following examples are only illustrative of the present invention and should not be construed as limiting the scope of the present invention.

Referring to fig. 1, the forest change remote sensing detection method based on the adaptive parameter genetic algorithm provided by the invention comprises the following steps:

step 1, acquiring two forest remote sensing images I at different moments₁And I₂Respectively preprocessing each forest remote sensing image to obtain two preprocessed forest remote sensing images I₁"and I₂"; the image size is M × N × B; m, N each represents I₁And I₂B represents the number of spectral bands of the image, wherein M is more than or equal to 20, N is more than or equal to 20, and B is more than or equal to 4;

step 1.1, reading forest remote sensing images of sentinels 2A at two different moments, wherein M is 350, N is 200, and B is 4;

and step 1.2, systematic and random radiation distortion or distortion is easily generated in the process of acquiring and transmitting the remote sensing image data, so that the distortion of the remote sensing image is caused, and the interpretation and interpretation of the remote sensing image are influenced. In order to eliminate or correct the image distortion caused by radiation error, the invention uses the forest remote sensing image I₁And I₂Inputting into remote sensing image processing software ENVI 5.2, and using radiometric calibration toolkit pair I of ENVI₁And I₂Carrying out radiation correction to obtain a radiation corrected forest remote sensing image I₁'、I₂‘；

Because factors such as atmosphere, illumination and the like can influence the reflectivity of ground objects, in order to acquire the real reflectivity of the ground objects, the invention utilizes the FLASH Atmospheric Correction toolkit in ENVI 5.2 to correct the forest remote sensing image I after radiation Correction₁'、I₂' atmospheric correction is carried out to obtain an atmospheric corrected forest remote sensing image I₁”、I₂", i.e. the pre-processed image.

Step 2, respectively calculating two preprocessed forest remote sensing images I by utilizing the reflectivity of the spectral band₁”、I₂"normalized vegetation index matrix NDVI₁、NDVI₂And making it be the corresponding vegetation index characteristic diagram SF₁、SF₂；

The vegetation index can combine the visible light and the near infrared wave band of the satellite according to the spectral characteristics of the plants, so that the effective measurement of the vegetation condition on the earth surface is realized, the vegetation and the soil are effectively distinguished, and the method is widely applied to the field of forest change detection. The normalized vegetation index NDVI is the best indicator factor for vegetation coverage, is sensitive to the change of soil background, eliminates the influence of terrain and shadow to a great extent, and has good vegetation extraction effect. The invention utilizes the reflectivity of the spectrum wave band of the remote sensing image to calculate the normalized vegetation index matrix NDVI of the forest remote sensing image after atmospheric correction, and the calculation formula is as follows:

where ρ is_NIRAnd ρ_REDRespectively representing infrared spectrum wave band reflectivity and red light spectrum wave band reflectivity of the forest remote sensing image, and enabling each preprocessed forest remote sensing image I₁”、I₂"the obtained NDVI matrix is the vegetation index characteristic diagram SF corresponding to the forest image₁、SF₂The image size is M × N, specifically 350 × 200.

Step 3, vegetation index characteristic diagram SF₁And SF₂Respectively carrying out median filtering to correspondingly obtain denoised feature maps SF₁'、SF₂'; denoising feature map SF by using difference method₁'、SF₂' construct vegetation index disparity map DI;

3.1, using median filtering algorithm to two vegetation index characteristic maps SF₁、SF₂De-noising to obtain de-noised vegetation index characteristic diagram SF₁'、SF₂’；

3.2, using difference method to denoise the characteristic diagram SF₁'、SF₂' construction of a vegetation index Difference map DI, i.e. two vegetation index signatures SF₁、SF₂And (4) making a difference on the pixel values of the corresponding positions, taking the difference value as the pixel value of the position corresponding to the DI (x, y), and traversing all the pixel points to obtain a difference map DI.

DI(x,y)＝SF₁'(x,y)-SF₂'(x,y)。

Step 4, carrying out significance detection on the vegetation index difference map DI to obtain a significance difference map DS;

4.1, in order to reduce the search space of classification calculation, the vegetation index difference map DI of the invention is subjected to bottom-up image significance detection based on global contrast. The method calculates the global contrast of each pixel on the whole image, takes the sum of Euclidean distances of the pixel and other pixels in the image on the color as the significance characteristic of the pixel, and the calculation formula is as follows:

wherein Sam (x, y) is a saliency feature of pixel (x, y), and (x, y) ≠ (i, j); sam (x, y) is then normalized to [0,255 ]. Since the vegetation index difference map DI is a single-channel image, the significant feature value is the gray value of the pixel.

4.2, generating a binaryzation significance difference map S by using an OTSU (maximum inter-class variance) image segmentation algorithm on the significance map Sam, wherein the pixel value of a non-significant region is 0, and the pixel value of a significant region is 1; and then generating a significance difference map DS by using the binarized significance map S and a vegetation index difference map DI:

DS(x,y)＝S(x,y)·DI(x,y)。

step 5, pre-classifying the significant difference image DS by using a fuzzy C-means clustering algorithm, dividing pixels in the DS into a variation class, an unchanged class and an uncertain class, and correspondingly obtaining a variation class pixel set, an unchanged class pixel set and an uncertain class pixel set;

5.1, setting an objective function of a fuzzy C-means clustering algorithm as follows:

wherein c is the number of categories, n is the total number of pixels of the significant difference map DS, u_ijRepresentative pixel x_jThe degree of membership belonging to the ith cluster,value of [0,1]And is andm represents a blurring weight coefficient, d_ij＝||x_j-v_i| l, representing pixel x_jAnd the clustering center v_iThe euclidean distance of (c).

In the specific operation, the invention sets c to 3, m to 2 and the threshold epsilon to 10^-5The maximum number of iterations T is 10³(ii) a Initializing the iteration number t to 1, and randomly initializing three cluster centers v₁、v₂、v₃；

5.2, calculating a new clustering center and membership function:

if the current iteration algebra t is 1, directly updating the membership degree of each pixel according to the pixel value of the clustering center; and if the current iteration algebra t is more than or equal to 2, updating the clustering center by using the membership matrix obtained by the previous generation, and updating the membership matrix according to the clustering center. The concrete formula is as follows:

wherein k is 1,2, 3;representing the ith cluster center in the t iteration,representing pixel x in the t-th iteration_jMembership belonging to the ith cluster;

5.3, ifAnd T is less than or equal to T, making T equal to T +1, and returning to the step 5.2; otherwise, the iteration is terminated and each pixel is assigned its final membershipCategory label: for pixel x_jIf arg_k{max{u_kj}}＝arg_k{max{v_k} and max { u }_kjIf the pixel is equal to or larger than 0.9, the pixel belongs to the variation class; arg_k{max{v_kDenotes the class k, arg with the largest cluster center pixel value_k{max{u_kjDenotes pixel x_jThe category corresponding to the maximum membership degree of (c); if arg_k{max{u_kj}}＝arg_k{min{v_k} and max { u }_kjIf the pixel is not less than 0.9, the pixel belongs to the unchanged class; otherwise, the pixel belongs to the uncertain class. The pixels of the determined classes, i.e. the changed and unchanged classes, are not calculated again in the algorithm after step 6.

Step 6, optimizing the class labels of the pixels in the uncertain class pixel set by using a genetic algorithm based on self-adaptive parameters to obtain optimized class labels corresponding to the uncertain classes; and generating a final change graph according to the optimized class labels.

6.1, population initialization:

setting the population size to be Num equal to 40, namely 40 individuals coded into a binary two-dimensional matrix in the population, wherein the size of an individual matrix Ind is the same as that of a significance difference map DS, namely 350 multiplied by 200; initializing all the individual matrixes Ind according to the class labels obtained in the step 5, namely modifying the pixel values of the positions corresponding to the changed pixel sets into 1, modifying the pixel values of the positions corresponding to the unchanged pixel sets into 0, and randomly coding the pixel values of the positions corresponding to the uncertain pixel sets into 0 or 1; and establishing a position index of the uncertain pixels in the individual matrix, and setting the iteration number k 'as 1 and the threshold epsilon' as 10^-2The maximum number of iterations K is 10⁵；

6.2, calculating individual fitness:

the probability of selecting an individual to perform a genetic operation is determined based on the fitness value of the individual to the objective function. Different from the traditional genetic algorithm, the improved genetic algorithm introduces the spatial information OF the pixels, designs an objective function OF with neighborhood factors, and has the following calculation formula:

wherein, r is a category identifier, r is 0,1, r is 0 corresponding to the unchanged category, and r is 1 corresponding to the changed category; c. C_rAll pixels representing the corresponding class in the salient region, n_rAnd m_rRepresenting the statistical number and average gray value, N, of pixels of the corresponding category_qRepresenting the central pixel x_j3 x 3 neighborhood N_jWherein q is 1,2,3, 9, l_qIs N_qAnd a central pixel x_jIs the Euclidean distance of L is the pixel x_jDistance weights to its surrounding pixels; DS (direct sequence)_jRepresenting the center pixel x in the significant difference map DS_jAnd (4) pixels corresponding to the positions.

After the objective function of each individual is obtained, the Fitness Fitness of each individual is obtained by taking the reciprocal of the objective function:

6.3, performing elite selection and cross operation on the population:

if the current iteration algebra k' is 1, performing genetic operation on all individuals; if the current iteration algebra k 'is more than or equal to 2, the current iteration individual is an offspring individual, the k' -1 generation individual is a parent individual, then, Num individuals are selected from high to low according to the fitness values of all the individuals of the offspring and the parent to execute genetic operation, namely 40 individuals are selected, the individual with the highest fitness value is selected as the best individual of the current iteration algebra, the best fitness value is stored in a matrix Maxfit, and the unselected individuals are eliminated in the evolution. And then performing cross operation on the selected individuals for executing the genetic operation, which specifically comprises the following steps: firstly, grouping all individuals performing genetic operation pairwise to form a plurality of groups to be crossed(ii) an individual; then, a matrix A with the same size as the individual is initialized randomly between (0,1), elements of the matrix A are traversed, and when the value of a certain element is smaller than the preset cross probability p_cWhen the value is 0.8, marking the position of the element; then exchanging elements in each group of individuals to be crossed, which are the same as the marked positions, and keeping other positions unchanged to form a crossed population;

6.4, calculating the variation probability of each individual in the crossed population pixel by pixel:

for each individual in the crossed population, traversing the individual matrix and aiming at a target pixel x_j3 x 3 neighborhood L_jEach neighborhood pixel L in_j(s), s 1,2,3, 9, whose membership u belongs to the class r is calculated_sr：

Wherein r is 0, 1: m is_rRepresenting the average gray value of each category of pixels obtained in the step 6.2);

according to the degree of membership u_srIs a neighborhood pixel L_j(s) assigning a class label if u_s0≥u_s1Then c is_sNot equal to 0, otherwise, c_s1 is ═ 1; thus, the target pixel x_jThe mutation probability of (c) is:

wherein d is_sIs a neighborhood pixel L_j(s) to center pixel point x_jSpatial distance of (B)_jIs x_jA binary coded value of;

if the obtained mutation probability p (x)_j) If the standard variation probability p is greater than the preset standard variation probability 0.2, the code of the position is changed, namely 0 is changed into 1 or 1 is changed into 0, otherwise, the code value is not changed; after traversing each individual matrix, obtaining a varied population, and taking the varied population as a progeny population of the previous generation; calculating the fitness of each individual in the filial generation population and selecting the fitnessThe most suitable fitness of the offspring is taken as the maximum fitness;

6.5, if the difference of the best fitness obtained by two adjacent iteration algebras is larger than a set threshold value, namely | Maxfit_k'+1-Maxfit_k'If | ≧ epsilon 'and K' is less than or equal to K, making K '═ K' +1, returning to step 6.3, otherwise, outputting the individual with the highest fitness of the iteration;

and 6.6, supplementing the class labels of the pixels corresponding to the changed classes and the unchanged classes obtained in the step 5 to the output optimal individuals, and outputting the final binary change map with the labels.

Simulation experiment

The correctness and effectiveness of the invention are further illustrated by the simulation data processing result.

Simulation content: the two multispectral forest images shown in fig. 2(a) and 2(b) are subjected to change detection by using the original genetic algorithm and the algorithm of the present invention, and the results are shown in fig. 2(c) and 2(d), wherein fig. 2(c) is a change map obtained by the original genetic algorithm, and fig. 2(d) is a change map obtained by the algorithm of the present invention.

Finally, the average classification accuracy was counted, and the results are shown in table 1.

TABLE 1 detection accuracy of original genetic Algorithm and the Algorithm of the present invention

Algorithm	Detection accuracy
		Primitive genetic algorithm	69.5％
Algorithm of the invention	82.2％

As can be seen from FIGS. 2(c) and 2(d), the method of the invention can realize accurate ground feature classification on the remote sensing image, can eliminate interference factors in forest ground features in a limited way, has stronger robustness and realizes accurate detection. And as can be seen from table 1, the present invention has higher detection accuracy.

According to the invention, the NDVI spectral characteristic diagram of the forest remote sensing image is used as a change detection to-be-divided image, and denoising and significance detection preprocessing are carried out on the change detection to-be-divided image. In order to further reduce the search space of the change detection algorithm, the invention uses fuzzy C-means clustering, assigns class labels to the pixels with the membership degree greater than the threshold value according to the gray value, and carries out genetic algorithm classification based on self-adaptive parameters on the pixels with the membership degree less than the threshold value.

The method is different from the traditional genetic algorithm in that the spatial neighborhood information of the image is introduced into the objective function and the variation probability, the genetic operator parameters are adaptively modified, the problems of poor noise immunity and huge search space of the traditional genetic algorithm in classification modeling are solved, noise suppression and detail retention are effectively balanced, the evolution is quickly converged towards the direction of the optimal solution, the detection precision of the algorithm is greatly improved, the variation information of forest regions is more accurately acquired, and the detection capability of forest land variation is enhanced.

Although the present invention has been described in detail in this specification with reference to specific embodiments and illustrative embodiments, it will be apparent to those skilled in the art that modifications and improvements can be made thereto based on the present invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.