WO2022258077A2 - 基于ii型模糊粗糙模型的遥感影像特征离散化方法、装置、存储介质、计算机设备 - Google Patents

基于ii型模糊粗糙模型的遥感影像特征离散化方法、装置、存储介质、计算机设备 Download PDF

Info

Publication number
WO2022258077A2
WO2022258077A2 PCT/CN2022/105555 CN2022105555W WO2022258077A2 WO 2022258077 A2 WO2022258077 A2 WO 2022258077A2 CN 2022105555 W CN2022105555 W CN 2022105555W WO 2022258077 A2 WO2022258077 A2 WO 2022258077A2
Authority
WO
WIPO (PCT)
Prior art keywords
remote sensing
sensing image
fuzzy
membership degree
type
Prior art date
Application number
PCT/CN2022/105555
Other languages
English (en)
French (fr)
Other versions
WO2022258077A3 (zh
Inventor
陈琼
黄小猛
Original Assignee
清华大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 清华大学 filed Critical 清华大学
Publication of WO2022258077A2 publication Critical patent/WO2022258077A2/zh
Publication of WO2022258077A3 publication Critical patent/WO2022258077A3/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/048—Fuzzy inferencing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/02—Computing arrangements based on specific mathematical models using fuzzy logic
    • G06N7/023—Learning or tuning the parameters of a fuzzy system
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing

Definitions

  • the present application relates to the field of remote sensing image feature extraction, in particular to a method, device, computer-readable storage medium, computer equipment, and computer program for discretizing remote sensing image features based on a Type II fuzzy rough model.
  • the feature discretization algorithm of remote sensing images is usually based on the assumption that a sample only belongs to a single category, which cannot describe the uncertainty caused by mixed pixels.
  • the sub-membership degree of mixed pixels is defined as a constant.
  • the fuzzy rough model quantifies the uncertainty information by introducing the membership degree of the pixel to each category, the decomposition model of the mixed pixel has a large error, which makes it inconsistent with the distribution information of the data and cannot accurately describe the uncertainty of the data. , resulting in a decrease in data accuracy.
  • the technical problem to be solved in this application is to overcome the defect that the feature discretization algorithm in the prior art cannot accurately quantify and evaluate the uncertainty caused by mixed pixels, so as to provide a remote sensing image feature based on type II fuzzy rough model Discretization method, device, computer readable storage medium, computer equipment, computer program.
  • the embodiment of the present application provides a method for discretizing remote sensing image features based on type II fuzzy rough model, including the following steps: acquiring target remote sensing image data, extracting mixed pixels from the target remote sensing image data, each mixed Each pixel contains the spectral response characteristics of multiple object types; determine the primary membership degree of each mixed pixel corresponding to each object type according to the mixed pixel; calculate the secondary membership degree of each mixed pixel attributable to each object type according to the primary membership degree ; According to the primary membership degree and secondary membership degree, determine the type II fuzzy rough set of each object type; carry out feature discretization processing on the target remote sensing image data, and obtain the optimal discretization result.
  • determining the main membership degree of each mixed pixel corresponding to each object type according to the mixed pixel includes: iteratively calculating the fuzzy mean vector and fuzzy covariance matrix of the preset fuzzy segmentation matrix, and the preset fuzzy segmentation matrix is composed of the mixed pixel The composition of the membership degree corresponding to each object type; according to the fuzzy segmentation matrix when the iterative calculation meets the iteration termination condition, determine the main membership degree of each mixed pixel corresponding to each object type.
  • the fuzzy segmentation matrix when the iterative calculation meets the iteration termination condition, determine the primary membership degree of each mixed pixel corresponding to each object type, including: according to the fuzzy segmentation matrix, determine the corresponding abundance of each mixed pixel, and Abundance is used as the primary membership degree of each mixed pixel corresponding to each object type.
  • calculate the secondary membership degree of each mixed pixel belonging to each object type according to the primary membership degree including: determine the hard segmentation matrix according to the fuzzy segmentation matrix when the iterative calculation meets the iteration termination condition; determine the attribution according to the hard segmentation matrix A set composed of pixels of each object type; calculate the upper approximation, lower approximation, positive domain, negative domain, and boundary domain of the set in the approximate space; determine each The mixed pixel belongs to the sub-subordination degree of each object type.
  • perform feature discretization processing on the remote sensing image data of the target to obtain the optimal discretization result including: obtaining the initial breakpoint set of mixed pixels from the remote sensing image data; initializing the target based on the number of breakpoints in the initial breakpoint set Remote sensing image data population; execute the genetic algorithm iteratively on the individual target remote sensing image data population to determine the optimal discretization result; wherein, the discretization scheme corresponding to the initialized target remote sensing image data population is the initial discretization scheme, each Population individuals correspond to a discretization result.
  • the genetic algorithm is iteratively performed on the individual of the target remote sensing image data population to determine the optimal discretization result, including: determining the fuzzy relationship between the mixed pixels based on the Euclidean distance between the mixed pixels; according to the fuzzy relationship, calculating II The average approximation precision of the type fuzzy rough set; according to the number of breakpoints in the initial breakpoint set, determine the reduction range of the number of breakpoints corresponding to the target remote sensing image data population; according to the reduction range of the number of breakpoints and the average approximation precision, determine The fitness function of type II fuzzy rough set; according to the fitness function of type II fuzzy rough set, the fitness value of type II fuzzy rough set is determined, and the individual of each target remote sensing image data population corresponding to the optimal fitness value is used as Optimal discretization results.
  • the fitness function of type II fuzzy rough set is expressed by the following formula:
  • ⁇ and ⁇ are the weight coefficients
  • is the magnitude of the reduction in the number of breakpoints
  • is the average approximate precision
  • an embodiment of the present application provides a remote sensing image feature discretization device based on a type II fuzzy rough model, including: a mixed pixel extraction unit configured to acquire target remote sensing image data, and extract from the target remote sensing image data Extract mixed pixels, each mixed pixel contains the spectral response characteristics of multiple object types; the primary membership degree determination unit is configured to determine the primary membership degree of each mixed pixel corresponding to each object type according to the mixed pixel; the secondary membership degree The degree determination unit is configured to calculate the secondary membership degree of each mixed pixel belonging to each object type according to the primary membership degree; the fuzzy rough set determination unit is configured to determine the II of each object type according to the primary membership degree and secondary membership degree. type fuzzy rough set; the optimal discretization result determination unit is configured to perform feature discretization processing on the target remote sensing image data to obtain the optimal discretization result.
  • an embodiment of the present application provides a non-transitory computer-readable storage medium.
  • the non-transitory computer-readable storage medium stores computer instructions.
  • any one of the first aspect The discretization method of remote sensing image features based on type II fuzzy rough model described in the implementation mode.
  • an embodiment of the present application provides a computer device, including at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, The instructions are executed by at least one processor, so as to execute the method for discretizing remote sensing image features based on a type II fuzzy rough model as described in any implementation manner of the first aspect.
  • an embodiment of the present application provides a computer program, the program is used to enable a processor to execute the method for discretizing remote sensing image features based on a Type II fuzzy rough model as described in any implementation manner of the first aspect.
  • the present application provides a remote sensing image feature discretization method, device, computer readable storage medium, computer equipment, and computer program based on a type II fuzzy rough model. Extract mixed pixels, each mixed pixel contains the spectral response characteristics of various object types; determine the main membership degree of each mixed pixel corresponding to each object type according to the mixed pixel; calculate the belongingness of each mixed pixel according to the main membership degree The secondary membership degree of each object type; according to the primary membership degree and secondary membership degree, determine the type II fuzzy rough set of each object type; perform feature discretization processing on the target remote sensing image data, and obtain the optimal discretization result.
  • the embodiment of the present application combines rough sets and fuzzy sets, uses the primary membership degree and secondary membership degree corresponding to the mixed pixel to describe the fuzzy components in the discretization process of remote sensing image features, uses the primary membership degree to fuzzy the discretization process, and uses the secondary membership degree
  • the membership degree further blurs the main membership degree, so as to accurately quantify and evaluate the uncertainty of mixed pixels, and obtain more accurate discretization results.
  • Fig. 1 is the flow chart of a specific example of the remote sensing image feature discretization method based on type II fuzzy rough model in embodiment 1 of the present application;
  • Fig. 2 is a specific example analysis diagram of the discretization method of remote sensing image features based on type II fuzzy rough model in embodiment 1 of the present application;
  • FIG. 3 is a structural example diagram of a remote sensing image feature discretization device based on a type II fuzzy rough model in Embodiment 2 of the present application;
  • FIG. 4 is a structural example diagram of a computer device in Embodiment 4 of the present application.
  • the fuzzy-rough model is a more powerful uncertainty data analysis model than fuzzy sets and rough sets. Fuzzy set is introduced on the basis of rough set, and the correlation between samples is described by using similarity relation instead of equivalence relation of rough set.
  • the type II fuzzy rough model can provide more accurate uncertainty analysis capabilities.
  • the type II fuzzy rough model fuzzifies the membership function value of the fuzzy set again, so that it can describe the fuzzy phenomenon more profoundly.
  • exp refers to the exponential function in advanced mathematics with the natural constant e as the base.
  • inf means the infimum, which is the largest lower bound of a set.
  • sup means the supremum, which is the least upper bound of a set.
  • This embodiment provides a method for discretizing remote sensing image features based on type II fuzzy rough model, as shown in Figure 1, including the following steps:
  • S11 Obtain remote sensing image data of the target, and extract mixed pixels from the remote sensing image data of the target, and each mixed pixel contains spectral response characteristics of various ground object types.
  • the components of the mixed pixel spectral signal are called endmembers, and each endmember corresponds to a spectral response characteristic of a ground object type.
  • S12 Determine the primary membership degree of each mixed pixel corresponding to each object type according to the mixed pixel.
  • the main membership degree of each corresponding surface object type according to the mixed pixel is to iteratively calculate the fuzzy segmentation matrix; when the iterative calculation meets the iteration termination condition, determine the corresponding abundance of each mixed pixel, and calculate the abundance As the primary membership degree of each mixed pixel corresponding to each object type.
  • the abundance corresponding to each mixed pixel refers to the abundance of the end members of the mixed pixel.
  • the fuzzy segmentation matrix is composed of the membership degree of each mixed pixel to the number of categories of the classification scheme.
  • the fuzzy mean value vector and fuzzy covariance matrix in the fuzzy partition matrix can be expressed by the above membership degree.
  • calculating the secondary membership degree of each mixed pixel belonging to each object type refers to determining the secondary membership degree according to the distribution of the mixed pixel in the boundary area of the rough set.
  • the distribution of mixed pixels in the rough set boundary area includes determining the set of pixels belonging to each object type; calculating the distribution area of the set in the approximate space to determine the sub-subordination degree of mixed pixels belonging to each object type.
  • the distribution area of the set in the approximate space includes: upper approximation, lower approximation, positive domain, negative domain, and boundary domain of the aggregate in the approximate space.
  • the approximation space refers to the rough approximation space (U,T), where U represents the set of mixed pixels, and T represents the number of bands of the remote sensing image.
  • S14 According to the primary membership degree and secondary membership degree, determine the type II fuzzy rough set of each object type.
  • the remote sensing image feature dispersion is described by the primary membership degree and the secondary membership degree.
  • the fuzzy components in the process of fuzzy discretization are fuzzy with the primary membership degree, and the primary membership degree is further fuzzy with the secondary membership degree.
  • the type II fuzzy rough set of each object type is determined. .
  • S15 Perform feature discretization processing on the target remote sensing image data to obtain an optimal discretization result.
  • discretization is to adopt a specific method to divide continuous features into multiple subintervals, and associate multiple subintervals with candidate breakpoints. Therefore, the feature discretization of target remote sensing images can be regarded as the selection of candidate breakpoints.
  • the process of discretizing the features of the target remote sensing image data to obtain the optimal discretization results refers to iteratively selecting candidate breakpoints through the genetic algorithm; and through the reduction of the number of candidate breakpoints in each iteration process and the average Approximate accuracy, determine the fitness function of individuals in the population; evaluate the discretization results with the determined fitness function of individuals in the population, and obtain the optimal discretization results.
  • the determination of the initial discretization scheme is determined by obtaining the initial breakpoint set of the mixed pixels in the remote sensing image data.
  • the present application provides a method for discretizing remote sensing image features based on type II fuzzy rough model.
  • Spectral response characteristics of object types determine the primary membership degree of each mixed pixel corresponding to each object type according to the mixed pixel; calculate the secondary membership degree of each mixed pixel belonging to each object type according to the primary membership degree; The degree of membership determines the type II fuzzy rough set of each object type; the feature discretization process is performed on the target remote sensing image data to obtain the optimal discretization result.
  • the embodiment of the present application combines rough sets and fuzzy sets, uses the primary membership degree and secondary membership degree corresponding to the mixed pixel to describe the fuzzy components in the discretization process of remote sensing image features, uses the primary membership degree to fuzzy the discretization process, and uses the secondary membership degree
  • the membership degree further blurs the main membership degree, so as to accurately quantify and evaluate the uncertainty of mixed pixels, and obtain more accurate discretization results.
  • step S12 determining the primary membership degree of each mixed pixel corresponding to each object type according to the mixed pixel, including:
  • the preset fuzzy segmentation matrix can be expressed as follows:
  • Fs(Xk) represents the membership degree of the k-th pixel Xk in U to the category s, s ⁇ 1,2,...,g ⁇ , g represents the number of categories, k ⁇ 1,2,...,n ⁇ , n represents the number of pixels in U.
  • Fs(Xk) satisfies: 0 ⁇ Fs(Xk) ⁇ 1, q ⁇ 1,2,...,g ⁇ .
  • fuzzy mean vector of the preset fuzzy segmentation matrix can be expressed by the following formula:
  • w represents the weight, w is greater than or equal to 1, and xik represents the pixel value of the k-th pixel on the i-th band.
  • fuzzy covariance matrix of the preset fuzzy segmentation matrix can be expressed as follows:
  • ⁇ mms represents fuzzy covariance, j ⁇ 1,2,...,m ⁇ ,
  • iteratively calculating the fuzzy mean vector and the fuzzy covariance matrix of the preset fuzzy segmentation matrix includes determining the preset fuzzy segmentation matrix from the fuzzy mean vector and the fuzzy covariance matrix.
  • the preset fuzzy segmentation matrix is composed of the membership degree of the pixel to the category, and the membership degree of the pixel to the category can be determined by the fuzzy mean vector and the fuzzy covariance matrix.
  • the membership degree of the pixel to the category can be calculated according to the following formula Express:
  • P ⁇ (s) is the prior probability of the occurrence of the sth category
  • the iteration termination condition can be expressed by the following formula:
  • ⁇ is the number of iteration steps of the algorithm
  • ⁇ is the error threshold
  • determining the primary membership degree of each mixed pixel corresponding to each object type includes: according to the fuzzy segmentation matrix, determining each mixed pixel The corresponding abundance is used as the main membership degree of each mixed pixel corresponding to each object type.
  • the determination of the primary membership degree of each mixed pixel corresponding to each object type is related to the weight and the abundance of each end member in the mixed pixel.
  • e is a natural number.
  • the primary membership degree of each mixed pixel corresponding to each object type can be expressed according to the following formula:
  • Ps(x) represents the abundance corresponding to the mixed pixel x
  • Jx represents the value range of the primary membership degree
  • u represents a primary membership degree
  • e determines the number of weights in w.
  • the weight w not only controls the convexity and concaveness of the discretization results, but also controls the sharing degree of mixed pixels among various types. The research results show that when the weight is 2, the effect of clustering can reach the best with a greater probability. If the value of e is greater than 1, different weights will be introduced, and the results of fuzzy clustering can be considered more comprehensively, but the clustering quality brought by the introduction of weights cannot be guaranteed, making the discretization results error-prone and unstable.
  • the value of e is 1 and the value of w is 2. at this time, It should be understood that the values of w and e include, but are not limited to, the methods described in the embodiments, as long as the values of w and e can be used to ensure the stability of the discretization result and reduce the complexity.
  • the fuzzy partition matrix is determined by iteratively calculating the fuzzy mean vector and the fuzzy partition matrix.
  • the abundance corresponding to the mixed pixel is determined by the fuzzy segmentation matrix, and the abundance is used as the main membership degree of the mixed pixel corresponding to each object type. In this way, the process of discretization is fuzzified by the degree of master membership.
  • the secondary membership degree of each mixed pixel belonging to each object type is calculated according to the primary membership degree, including:
  • the determination of the hard segmentation matrix is to modify the maximum value of each column in the fuzzy segmentation matrix when the iteration termination condition is met to 1, and modify the other values in the corresponding columns to 0, thereby completing the determination of the hard segmentation matrix.
  • the set of pixels belonging to each object type can be expressed according to the following formula:
  • Xs represents the set of pixels belonging to the sth type of ground object in the mixed pixel set
  • Cs(x) is the membership degree of the pixel x in the hard segmentation matrix C to the sth category.
  • the hard segmentation matrix has determined the corresponding position with a membership degree value of 1 in each column, and judges whether the mixed pixel in the mixed pixel set U belongs to According to each object type, the set of pixels belonging to each object type can be determined according to the hard segmentation matrix.
  • calculating the upper approximation, lower approximation, positive domain, negative domain, and boundary domain of the set Xs composed of pixels belonging to each object type in the approximate space includes: determining the equivalence class set of the mixed pixel set; The equivalence class set calculates the upper approximation, lower approximation, positive domain, negative domain, and boundary domain of the set composed of pixels belonging to each object type in the approximate space.
  • the equivalence class set of the mixed pixel set can be expressed by the following formula:
  • the upper approximation of Xs in the approximation space can be calculated according to the following formula:
  • T*(Xs) represents the upper approximation of Xs in the approximation space.
  • T*(Xs) represents the lower approximation of the set in the approximation space.
  • the positive domain of Xs in the approximate space can be calculated according to the following formula:
  • POST(Xs) represents the positive domain of the set in the approximate space.
  • the negative domain of Xs in the approximate space can be calculated according to the following formula:
  • NGTT(Xs) represents the negative domain of the set in the approximate space.
  • boundary domain of Xs in the approximate space can be calculated according to the following formula:
  • BNT(Xs) represents the boundary domain of the collection in the approximate space.
  • the secondary membership degree of each mixed pixel belonging to each object type including: according to the upper approximation, lower approximation, positive domain, negative domain, and boundary domain , determine the probability that the pixel belongs to the set composed of pixels of each object type; according to the determined probability that each pixel belongs to the set composed of pixels of each object type, determine the secondary subordination of each mixed pixel to belong to each object type Spend.
  • determining the probability that a pixel belongs to a set composed of pixels of each object type means that according to the upper approximation, lower approximation, positive domain, negative domain,
  • the boundary domain determines the distribution of the pixels; with the determined distribution, according to the Bayesian theorem, determine the probability that the pixel belongs to the set of pixels of each object type.
  • POST(Xs) is a set of pixels belonging to Xs in the mixed pixel set
  • NGTT(Xs) is a set of pixels not belonging to Xs in the mixed pixel set
  • BNT(Xs) is In the mixed pixel set, it is a set composed of pixels that cannot definitely belong to Xs. Therefore, BNT(Xs) is the uncertainty domain of mixed pixels.
  • Fig. 2 an analysis diagram of a specific example of the discretization method of remote sensing image features based on the type II fuzzy rough model in the embodiment of the present application, each rectangle in the figure represents an equivalence class in U
  • the circular area corresponding to the reference number 22 represents Xs
  • the octagonal area corresponding to the reference number 24 represents T* (Xs)
  • the rectangular area corresponding to the reference number 23 represents T* (Xs) or POST (Xs ).
  • the area outside the octagon corresponding to reference numeral 24 represents NGTT (Xs)
  • the area inside the octagon corresponding to reference numeral 24 and outside the rectangle corresponding to reference numeral 23 represents BNT (Xs).
  • determine the probability that the pixel belongs to the set formed by the pixels of each object type including: the probability that the pixel belongs to the set formed by the pixels of each object type can be Expressed by the following formula:
  • E[i] represents the i-th equivalence class on the pixel set
  • G[s] represents the s-th category
  • E[i]) represents the image of category s in the equivalence class i
  • G[s]) indicates the proportion of pixels belonging to the equivalent class i among all the pixels of category s
  • P(E[i]) indicates the proportion of P(G[s]) represents the probability that a pixel of category s appears in U.
  • each pixel belongs to the set of pixels of each object type determines the sub-subordination degree of each mixed pixel belonging to each object type, including: each mixed pixel belongs to each object type.
  • the degree of membership can be expressed by the following formula:
  • the upper approximation, the lower approximation, the positive domain, the negative domain, and the boundary domain of the set in the approximation space are calculated by determining the set of pixels of each object type; according to the upper approximation, the lower Approximate, positive domain, negative domain, boundary domain, determine the probability that the pixel belongs to the set of pixels of each object type; according to the determined probability that the pixel belongs to the set of pixels of each object type, determine the probability of each mixed The sub-subordination degree of the pixel belonging to each object type.
  • the discretization scheme is equivalent to dividing the mixed pixel set into multiple equivalence classes, the elements in the same equivalence class have the same attribute value.
  • the mixed pixels in the same equivalence class have the same probability of belonging to each category, but for the mixed pixels in different equivalence classes, their probability of belonging to each category will be different.
  • the uncertainty brought by the difference is a further blurring of the primary membership degree. Therefore, the process of determining the secondary membership degree in the technical solution of the present application is equivalent to further blurring the primary membership degree through the secondary membership degree, so as to accurately quantify and evaluate the uncertainty of the mixed pixel.
  • step S14 for The type II fuzzy rough set of each object type is expressed by the following formula:
  • the feature discretization process is performed on the target remote sensing image data to obtain the optimal discretization result, including:
  • obtaining the initial breakpoint set of mixed pixels from the remote sensing image data refers to obtaining the initial breakpoint set from the input remote sensing image features.
  • the acquisition of the initial breakpoint set belongs to a relatively mature prior art, which will not be described in detail in this application.
  • discretization is to adopt a specific method to divide continuous features into multiple subintervals, and associate multiple subintervals with candidate breakpoints. Therefore, the feature discretization process of the target remote sensing image can be regarded as the selection of candidate breakpoints, and each discretization scheme corresponds to a division result on the mixed pixel set.
  • the initial target remote sensing image data population refers to the number of breakpoints in the initial breakpoint set as the individual length of the initial population, so as to complete the initialization of the target remote sensing image data population.
  • the length of the individual is the number of initial breakpoints, that is, the length of each population individual is 10 bits, corresponding to the initial breakpoint set 10 breakpoints.
  • the discretization scheme corresponding to the individual in the initial population is the initial discretization scheme
  • the discretization result corresponding to the initial discretization scheme is the initial discretization result
  • the number of individuals in the target remote sensing image data population is 50
  • the 50 individuals in the population will undergo selection, mutation , Crossover, that is, the evolution function of the genetic algorithm to update the population, so that the 50 individuals in the population of the next generation are different from the 50 individuals of the previous generation.
  • the genetic algorithm is iteratively executed on the individual of the target remote sensing image data population to determine the optimal discretization result, including:
  • fuzzy relationship between mixed pixels can be expressed by the following formula:
  • d(x,y) represents the Euclidean distance between x and y
  • xh and yh represent the pixel values of x and y in the h-band, respectively.
  • calculating the average approximation accuracy of the type II fuzzy rough set according to the fuzzy relationship includes: calculating the upper approximation and the lower approximation of the type II fuzzy rough set according to the fuzzy relationship; With the lower approximation, the average approximation accuracy of type II fuzzy rough sets is calculated.
  • type II fuzzy rough set can be expressed by the following formula:
  • u2 is one of the master membership degrees
  • Jy2 represents the value range of the corresponding membership degree
  • a(y2) represents the secondary membership degree of the mixed pixel y2, and respectively The minimum and maximum values of the primary membership degree of .
  • u1 is one of the master membership degrees
  • Jy1 represents the value range of the corresponding membership degree
  • a(y1) represents the secondary membership degree of the mixed pixel y1.
  • the average approximation precision of type II fuzzy rough set can be expressed by the following formula:
  • the magnitude of the reduction in the number of breakpoints corresponding to the individual of the target remote sensing image data population including: determining the length of the individual as the number of initial breakpoints; determining the length of the selected individual
  • the number is the number of breakpoints corresponding to the individual; the difference between the length of the individual and the number of individual breakpoints is determined to be the magnitude of reduction in the number of breakpoints corresponding to the individual of the target remote sensing image data population.
  • the length of the individual is the number of initial breakpoints, that is, the length of each population individual is 10 bits, corresponding to the initial breakpoint set 10 breakpoints.
  • Each bit in the binary code corresponds to a candidate breakpoint, and the values '1' and '0' represent that the breakpoint is selected and not selected, respectively.
  • determine the selected number of binary codes as the number of breakpoints of the corresponding individual that is, determine the number of binary bits with a value of '1' to obtain the number of breakpoints of the individual.
  • the magnitude of the reduction in the number of breakpoints refers to the amount of reduction in the number of breakpoints of population individuals in each iteration process.
  • the discretization effect is measured by comparing the reduction in the number of population breakpoints in the iterative process, and the greater the reduction in the number of breakpoints, the better the discretization effect.
  • the fitness function of type II fuzzy rough set is used to calculate the fitness value of each target remote sensing image data population individual. and pose.
  • the fitness function of type II fuzzy rough set is expressed by the following formula:
  • ⁇ and ⁇ are the weight coefficients
  • is the magnitude of the reduction in the number of breakpoints
  • is the average approximate precision
  • the selection of the weight coefficient is based on the actual working conditions, and the rationality of the weight setting is generally judged according to the characteristics of the data set and experimental observations, which is not specifically limited in this application.
  • multiple discretization schemes are used as population individuals in the genetic algorithm, through the evolution function of the genetic algorithm, iterative calculations are used to find the individual with the largest fitness value, and the individual with the largest fitness value is used as the optimal fitness value individual.
  • the discretization scheme corresponding to the optimal fitness value is the optimal discretization scheme.
  • the fitness function corresponding to the population individuals has 50 values.
  • these 50 individuals in the population undergo evolution to update the population.
  • the 50 individuals in the population of the next generation are different from the 50 individuals of the previous generation.
  • the global variable will record the individual with the highest fitness value among the 50 individuals.
  • the global variable is updated with the individual with a higher fitness value.
  • the fitness function of the type II fuzzy rough set is constructed by calculating the average approximation precision of the type II fuzzy rough set and determining the reduction magnitude of the number of breakpoints.
  • the individual with the optimal fitness value is determined by genetic algorithm, and the fuzzy components in the discretization process of remote sensing image features are described by the primary membership degree and secondary membership degree corresponding to the mixed pixel, and the discretization process is blurred by the primary membership degree, and The primary membership degree is further fuzzified with the secondary membership degree, and more accurate discretization results are obtained through the fitness function of the constructed type II fuzzy rough set.
  • This embodiment provides a remote sensing image feature discretization device based on a type II fuzzy rough model, as shown in Figure 3, which is a remote sensing image feature based on a type II fuzzy rough model provided by an optional embodiment of the present application
  • the connection diagram of the discretization device includes: a mixed pixel extraction unit 31 , a primary membership degree determination unit 32 , a secondary membership degree determination unit 33 , a fuzzy rough set determination unit 34 , and an optimal discretization result determination unit 35 .
  • the mixed pixel extraction unit 31 is configured to acquire the remote sensing image data of the target, and extract mixed pixels from the remote sensing image data of the target, and each mixed pixel contains spectral response characteristics of various types of ground objects.
  • step S11 refer to the relevant description of step S11 in any of the above method embodiments, which will not be repeated here.
  • the main membership degree determining unit 32 is configured to determine the main membership degree of each mixed pixel corresponding to each object type according to the mixed pixel. For details, refer to the relevant description of step S12 in any of the above method embodiments, and details are not repeated here.
  • the secondary membership degree determining unit 33 is configured to calculate the secondary membership degree of each mixed pixel belonging to each object type according to the primary membership degree. For details, reference may be made to the relevant description of step S13 in any of the above method embodiments, and details are not repeated here.
  • the fuzzy rough set determination unit 34 is configured to determine the type II fuzzy rough set of each object type according to the primary membership degree and the secondary membership degree. For details, refer to the relevant description of step S14 in any of the above method embodiments, and details are not repeated here.
  • the optimal discretization result determining unit 35 is configured to perform feature discretization processing on the target remote sensing image data to obtain an optimal discretization result. For details, refer to the relevant description of step S15 in any of the above method embodiments, and details are not repeated here.
  • the present application provides a remote sensing image feature discretization device based on a type II fuzzy rough model, which includes: a mixed pixel extraction unit 31 configured to acquire target remote sensing image data, and extract mixed pixel elements from the target remote sensing image data , each mixed pixel contains the spectral response characteristics of multiple object types.
  • the main membership degree determining unit 32 is configured to determine the main membership degree of each mixed pixel corresponding to each object type according to the mixed pixel.
  • the secondary membership degree determining unit 33 is configured to calculate the secondary membership degree of each mixed pixel belonging to each object type according to the primary membership degree.
  • the fuzzy rough set determination unit 34 is configured to determine the type II fuzzy rough set of each object type according to the primary membership degree and the secondary membership degree.
  • the optimal discretization result determining unit 35 is configured to perform feature discretization processing on the target remote sensing image data to obtain an optimal discretization result.
  • the embodiment of the present application combines rough sets and fuzzy sets, uses the primary membership degree and secondary membership degree corresponding to the mixed pixel to describe the fuzzy components in the discretization process of remote sensing image features, uses the primary membership degree to fuzzy the discretization process, and uses the secondary membership degree
  • the membership degree further blurs the main membership degree, so as to accurately quantify and evaluate the uncertainty of mixed pixels, and obtain more accurate discretization results.
  • the main membership degree determination unit 32 includes: an iterative calculation subunit and a main membership degree determination subunit.
  • an iterative calculation subunit and a main membership degree determination subunit.
  • the iterative calculation subunit is configured to iteratively calculate the fuzzy mean vector and the fuzzy covariance matrix of the preset fuzzy segmentation matrix, wherein the preset fuzzy segmentation matrix is composed of membership degrees of mixed pixels corresponding to various object types.
  • the main membership degree determination subunit is configured to determine the main membership degree of each mixed pixel corresponding to each object type according to the fuzzy segmentation matrix when the iterative calculation meets the iteration termination condition.
  • the sub-membership determination subunit including the abundance determination subunit, is configured to determine the abundance corresponding to each mixed pixel according to the fuzzy segmentation matrix, and use the abundance as each mixed image
  • the element corresponds to the primary membership degree of each object type. For details, please refer to the relevant description of determining the primary membership degree of each mixed pixel corresponding to each object type according to the iterative calculation of the fuzzy segmentation matrix when the iteration termination condition is met in any of the above method embodiments.
  • the secondary membership determination unit 33 includes: a hard segmentation matrix determination subunit, a pixel set determination subunit, an approximate space boundary determination subunit, and a secondary membership determination subunit.
  • a hard segmentation matrix determination subunit for details, please refer to the relevant description about calculating the secondary membership degree of each mixed pixel belonging to each object type according to the primary membership degree in any of the above method embodiments.
  • the hard partition matrix determination subunit is configured to determine the hard partition matrix according to the fuzzy partition matrix when the iterative calculation meets the iteration termination condition.
  • the pixel set determination subunit is configured to determine a set of pixels belonging to each object type according to the hard segmentation matrix.
  • the approximate space boundary determination subunit is configured to calculate the upper approximation, lower approximation, positive domain, negative domain, and boundary domain of the set in the approximate space.
  • the secondary membership determination subunit is configured to determine the secondary membership of each mixed pixel belonging to each object type according to the upper approximation, the lower approximation, the positive domain, the negative domain, and the boundary domain.
  • the optimal discretization result determination unit 35 includes an initial breakpoint set acquisition subunit, a population initialization subunit, and an optimal discretization result determination subunit.
  • an initial breakpoint set acquisition subunit for details, please refer to the related description about discretizing target remote sensing image data to obtain an optimal discretization result in any of the above method embodiments.
  • the initial breakpoint set acquisition subunit is configured to acquire an initial breakpoint set of mixed pixels from the remote sensing image data.
  • the population initialization subunit is configured to initialize the target remote sensing image data population based on the number of breakpoints in the initial breakpoint set.
  • the discretization scheme corresponding to the initialized target remote sensing image data population is the initial discretization scheme, and each population individual corresponds to a discretization result.
  • the optimal discretization result determination subunit is configured to iteratively execute the genetic algorithm on the individual iterations of the target remote sensing image data population to determine the optimal discretization result.
  • the subunit for determining the optimal discretization result includes: a subunit for determining the fuzzy relationship between pixels, a subunit for calculating, a subunit for determining the magnitude of the reduction in the number of breakpoints, and a subunit for determining the fitness function , the fitness value determines the subunit.
  • a subunit for determining the fuzzy relationship between pixels includes: a subunit for calculating, a subunit for determining the magnitude of the reduction in the number of breakpoints, and a subunit for determining the fitness function , the fitness value determines the subunit.
  • the fuzzy relationship determining subunit between the pixels is configured to determine the fuzzy relationship between the mixed pixels based on the Euclidean distance between the mixed pixels.
  • the calculation subunit is configured to calculate the average approximation precision of the type II fuzzy rough set according to the fuzzy relationship.
  • the subunit for determining the extent of reduction in the number of breakpoints is configured to determine the extent of reduction in the number of breakpoints corresponding to the individual target remote sensing image data population according to the number of breakpoints in the initial set of breakpoints.
  • the fitness function determination subunit is configured to determine the fitness function of the type II fuzzy rough set according to the reduction range of the number of breakpoints and the average approximation precision.
  • the fitness value determination subunit is configured to determine the fitness value of the type II fuzzy rough set according to the fitness function of the type II fuzzy rough set, and use the individual of each target remote sensing image data population corresponding to the optimal fitness value as Optimal Discretization Results.
  • the present application provides a remote sensing image feature discretization device based on a type II fuzzy rough model, which includes: a mixed pixel extraction unit 31 configured to acquire target remote sensing image data, and extract mixed pixel elements from the target remote sensing image data , each mixed pixel contains the spectral response characteristics of multiple object types.
  • the main membership degree determining unit 32 is configured to determine the main membership degree of each mixed pixel corresponding to each object type according to the mixed pixel.
  • the secondary membership degree determining unit 33 is configured to calculate the secondary membership degree of each mixed pixel belonging to each object type according to the primary membership degree.
  • the fuzzy rough set determination unit 34 is configured to determine the type II fuzzy rough set of each object type according to the primary membership degree and the secondary membership degree.
  • the optimal discretization result determining unit 35 is configured to perform feature discretization processing on the target remote sensing image data to obtain an optimal discretization result.
  • the embodiment of the present application combines rough sets and fuzzy sets, uses the primary membership degree and secondary membership degree corresponding to the mixed pixel to describe the fuzzy components in the discretization process of remote sensing image features, uses the primary membership degree to fuzzy the discretization process, and uses the secondary membership degree
  • the membership degree further blurs the main membership degree, so as to accurately quantify and evaluate the uncertainty of mixed pixels, and obtain more accurate discretization results.
  • An embodiment of the present application also provides a non-transitory computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the method described in any of the above method embodiments.
  • the storage medium can be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a flash memory (Flash Memory), a hard disk (Hard Disk) Disk Drive, abbreviation: HDD) or solid-state hard drive (Solid-State Drive, SSD) etc.;
  • the storage medium can also include the combination of above-mentioned types of memory.
  • FIG. 4 is a schematic structural diagram of a computer device provided in an optional embodiment of the present application.
  • the computer device may include at least one processor 41, at least A communication interface 42, at least one communication bus 43 and at least one memory 44, wherein the communication interface 42 may include a display screen (Display), a keyboard (Keyboard), and the optional communication interface 42 may also include a standard wired interface and a wireless interface.
  • the memory 44 may be a random access memory (Random Access Memory, ie, a volatile memory), or a non-volatile memory (non-volatile memory), such as at least one disk memory.
  • the memory 44 may also be at least one storage device located away from the aforementioned processor 41 .
  • the processor 41 can be combined with the device described in FIG. 3 , the memory 44 stores an application program, and the processor 41 invokes the program code stored in the memory 44 to execute the steps of the method described in any of the above method embodiments.
  • the communication bus 43 may be a peripheral component interconnect standard (peripheral component interconnect, PCI) bus or an extended industry standard architecture (extended industry standard architecture, EISA) bus or the like.
  • PCI peripheral component interconnect
  • EISA extended industry standard architecture
  • the communication bus 43 can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, only one thick line is used in FIG. 4 , but it does not mean that there is only one bus or one type of bus.
  • the memory 44 may include a volatile memory (volatile memory), such as a random-access memory (random-access memory, RAM); the memory may also include a non-volatile memory (non-volatile memory), such as a flash memory ( flash memory), a hard disk (hard disk drive, HDD) or a solid-state drive (solid-state drive, SSD); the memory 44 can also include a combination of the above-mentioned types of memory.
  • volatile memory such as a random-access memory (random-access memory, RAM)
  • non-volatile memory such as a flash memory ( flash memory), a hard disk (hard disk drive, HDD) or a solid-state drive (solid-state drive, SSD)
  • the memory 44 can also include a combination of the above-mentioned types of memory.
  • the processor 41 may be a central processing unit (central processing unit, CPU), a network processor (network processor, NP) or a combination of CPU and NP.
  • CPU central processing unit
  • NP network processor
  • the processor 41 may further include a hardware chip.
  • the aforementioned hardware chip may be an application-specific integrated circuit (application-specific integrated circuit, ASIC), a programmable logic device (programmable logic device, PLD) or a combination thereof.
  • the aforementioned PLD may be a complex programmable logic device (complex programmable logic device, CPLD), a field-programmable gate array (field-programmable gate array, FPGA), a general array logic (generic array logic, GAL) or any combination thereof.
  • memory 44 is also used to store program instructions.
  • the processor 41 may invoke program instructions to implement the method described in any embodiment of the present application.
  • An embodiment of the present application further provides a computer program, the program is configured to cause a processor to execute the steps of the method described in any of the above method embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Automation & Control Theory (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Fuzzy Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Algebra (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

本申请提供了一种基于II型模糊粗糙模型的遥感影像特征离散化方法、装置、计算机可读存储介质、计算机设备、计算机程序，该方法包括：获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征；根据混合像元确定各混合像元对应各地物类型的主隶属度；根据主隶属度计算各混合像元归属于各地物类型的次隶属度；根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集；对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。本申请实施例以混合像元对应的主隶属度和次隶属度描述遥感影像特征离散化过程中的模糊成分，以主隶属度模糊离散化过程，并以次隶属度将主隶属度进一步模糊化，准确量化混合像元的不确定性，获得精确离散化结果。

Description

基于II型模糊粗糙模型的遥感影像特征离散化方法、装置、存储介质、计算机设备
本申请要求在2022年07月04日提交中国专利局、申请号为202210776562.X、发明名称为“基于II型模糊粗糙模型的遥感影像特征离散化方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及遥感影像特征提取领域，具体涉及一种基于II型模糊粗糙模型的遥感影像特征离散化方法、装置、计算机可读存储介质、计算机设备、计算机程序。
背景技术
遥感作为一种先进的技术手段，已经广泛应用于经济和社会发展的各个领域。空间分辨率、时间分辨率、光谱分辨率以及辐射分辨率的逐渐提高，使得遥感数据具有明显的大数据特征。由于地物要素空间分布的多样性，穿插性和复杂性，遥感图像中每个像元的光谱信号记录着不同的土地覆盖类型，这些像元称为混合像元。特征离散化作为一种最有影响力的数据预处理技术在广泛应用于工业控制的知识发现和数据挖掘领域扮演着重要角色。它能够将连续特征转换成更接近知识层表示的离散特征，使得数据更易于理解，使用和解释，从而提升遥感数据处理的效率和适应那些需要离散型数据作为输入的学习算法。在相关技术中，为实现遥感影像特征离散化，遥感影像的特征离散化算法通常基于一个样本仅属于单一类别的假设，无法描述混合像元引起的不确定性。或者为了简化II型模糊集合的运算，将混合像元的次隶属度定义为常量。尽管模糊粗糙模型通过引入像元对各类别的隶属度来量化不确定性信息，但是混合像元的分解模型存在较大的误差，造成与数据的分布信息不服，无法准确的描述数据的不确定性，造成数据精度的下降。
发明内容
因此，本申请要解决的技术问题在于克服现有技术中的特征离散化算法无法准确量化和评估混合像元引起的不确定性的缺陷，从而提供一种基于II型模糊粗糙模型的遥感影像特征离散化方法、装置、计算机可读存储介质、计算机设备、计算机程序。
根据第一方面，本申请实施例提供了一种基于II型模糊粗糙模型的遥感影像特征离散化方法，包括以下步骤：获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征；根据混合像元确定各混合像元对应各地物类型的主隶属度；根据主隶属度计算各混合像元归属于各地物类型的次隶属度；根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集；对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。
可选地，根据混合像元确定各混合像元对应各地物类型的主隶属度，包括：迭代计算预设模糊分割矩阵的模糊均值矢量和模糊协方差矩阵，预设模糊分割矩阵由混合像元对应各地物类型的隶属度组成；根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各混合像元对应各地物类型的主隶属度。
可选地，根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各混合像元对应各地 物类型的主隶属度，包括：根据模糊分割矩阵，确定各混合像元对应的丰度，并将丰度作为各混合像元对应各地物类型的主隶属度。
可选地，根据主隶属度计算各混合像元归属于各地物类型的次隶属度，包括：根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定硬分割矩阵；根据硬分割矩阵，确定归属于各地物类型的像元构成的集合；计算集合在近似空间中的上近似、下近似、正域、负域、边界域；根据上近似、下近似、正域、负域、边界域确定各混合像元归属于各地物类型的次隶属度。
可选地，对目标遥感图像数据进行特征离散化处理，得到最优离散化结果，包括：从遥感图像数据中获取混合像元的初始断点集；基于初始断点集的断点数量初始化目标遥感图像数据种群；对目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果；其中，与初始化后的目标遥感图像数据种群相对应的离散化方案为初始离散化方案，每个种群个体对应一个离散化结果。
可选地，对目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果，包括：基于混合像元间的欧氏距离确定混合像元间的模糊关系；根据模糊关系，计算II型模糊粗糙集的平均近似精度；根据初始断点集的断点数量，确定与目标遥感图像数据种群个体相对应的断点数目减少的幅度；根据断点数目减少的幅度与平均近似精度，确定II型模糊粗糙集的适应度函数；根据II型模糊粗糙集的适应度函数，确定II型模糊粗糙集的适应度值，并以最优适应度值对应的各目标遥感图像数据种群的个体作为最优离散化结果。
可选地，II型模糊粗糙集的适应度函数通过如下公式表达：
Figure PCTCN2022105555-appb-000001
其中，α和β为权重系数，|D|为以断点数目减少的幅度，
Figure PCTCN2022105555-appb-000002
为平均近似精度。
根据第二方面，本申请实施例提供了一种基于II型模糊粗糙模型的遥感影像特征离散化装置，包括：混合像元提取单元，被配置为获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征；主隶属度确定单元，被配置为根据混合像元确定各混合像元对应各地物类型的主隶属度；次隶属度确定单元，被配置为根据主隶属度计算各混合像元归属于各地物类型的次隶属度；模糊粗糙集确定单元，被配置为根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集；最优离散化结果确定单元，被配置为对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。
根据第三方面，本申请实施例提供了一种非暂态计算机可读存储介质，非暂态计算机可读存储介质存储有计算机指令，计算机指令被处理器执行时，实现如第一方面任一实施方式所述的基于II型模糊粗糙模型的遥感影像特征离散化方法。
根据第四方面，本申请实施例提供了一种计算机设备，包括至少一个处理器；以及与至少一个处理器通信连接的存储器；其中，存储器存储有可被所述至少一个处理器执行的指令，指令被至少一个处理器执行，从而执行如第一方面任一实施方式所述的基于II型模糊粗糙模型的遥感影像特征离散化方法。
根据第五方面，本申请实施例提供了一种计算机程序，所述程序用于使处理器执行如第一方面任一实施方式所述的基于II型模糊粗糙模型的遥感影像特征离散化方法。
本申请技术方案，具有如下优点：
本申请提供的一种基于II型模糊粗糙模型的遥感影像特征离散化方法、装置、计算机 可读存储介质、计算机设备、计算机程序，该方法包括：获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征；根据混合像元确定各混合像元对应各地物类型的主隶属度；根据主隶属度计算各混合像元归属于各地物类型的次隶属度；根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集；对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。本申请实施例结合粗糙集与模糊集，以混合像元对应的主隶属度和次隶属度来描述遥感影像特征离散化过程中的模糊成分，以主隶属度模糊化离散化过程，并以次隶属度将主隶属度进一步模糊化，从而准确量化和评估混合像元的不确定性，获得更加精确地离散化结果。
附图说明
为了更清楚地说明本申请具体实施方式或现有技术中的技术方案，下面将对具体实施方式或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施方式，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例1中基于II型模糊粗糙模型的遥感影像特征离散化方法的一个具体示例的流程图；
图2为本申请实施例1中基于II型模糊粗糙模型的遥感影像特征离散化方法的一个具体示例分析图；
图3为本申请实施例2中基于II型模糊粗糙模型的遥感影像特征离散化装置的结构示例图；
图4为本申请实施例4中计算机设备的结构示例图。
具体实施方式
下面将结合附图对本申请的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
在本申请的描述中，需要说明的是，术语“中心”、“上”、“下”、“左”、“右”、“竖直”、“水平”、“内”、“外”等指示的方位或位置关系为基于附图所示的方位或位置关系，仅是为了便于描述本申请和简化描述，而不是指示或暗示所指的装置或元件必须具有特定的方位、以特定的方位构造和操作，因此不能理解为对本申请的限制。此外，术语“第一”、“第二”、“第三”仅用于描述目的，而不能理解为指示或暗示相对重要性。
在本申请的描述中，模糊粗糙模型是一个比模糊集和粗糙集更加强大的不确定性数据分析模型。在粗糙集的基础上引入模糊集，通过采用相似关系代替粗糙集的等价关系来描述样本之间的相关性。作为模糊粗糙模型的推广，II型模糊粗糙模型能够提供更准确的不确定性分析能力。II型模糊粗糙模型将模糊集合的隶属函数值再次进行模糊化，从而能够更深刻地描述模糊现象。
在本申请关于公式的描述中，exp是指高等数学里以自然常数e为底的指数函数。inf表示下确界，是一个集合的最大下界。sup表示上确界，是一个集合的最小上界。
此外，下面所描述的本申请不同实施方式中所涉及的技术特征只要彼此之间未构成冲突就可以相互结合。
实施例1
本实施例提供一种基于II型模糊粗糙模型的遥感影像特征离散化方法，如图1所示，包括如下步骤：
S11：获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征。
具体地，混合像元光谱信号的组成成分称为端元，每个端元对应一种地物类型的光谱响应特征。
S12：根据混合像元确定各混合像元对应各地物类型的主隶属度。
具体地，根据混合像元确定各对应地物类型的主隶属度是通过迭代计算模糊分割矩阵；在迭代计算满足迭代终止条件的情况下，确定各混合像元对应的丰度，并将丰度作为各混合像元对应各地物类型的主隶属度。其中，各混合像元对应的丰度是指混合像元的端元的丰度。
在实际应用中，模糊分割矩阵由各混合像元对于分类方案的类别数目的隶属度组成。模糊分割矩阵中的模糊均值矢量和模糊协方差矩阵可通过上述隶属度表示。
S13：根据主隶属度计算各混合像元归属于各地物类型的次隶属度。
具体地，计算各混合像元归属于各地物类型的次隶属度是指根据混合像元在粗糙集边界区域的分布情况确定次隶属度。混合像元在粗糙集边界区域的分布情况包括确定归属于各地物类型的像元构成的集合；计算集合在近似空间中的分布区域确定混合像元归属于各地物类型的次隶属度。其中，集合在近似空间的分布区域，包括：集合在近似空间中的上近似、下近似、正域、负域、边界域。
在实际应用中，近似空间是指粗糙近似空间(U,T)，其中U表示混合像元集合，T表示遥感影像的波段数量。
S14：根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集。
具体地，在确定了各混合像元对应各地物类型的主隶属度与各混合像元归属于各地物类型的次隶属度的过程，是以主隶属度和次隶属度来描述遥感影像特征离散化过程中的模糊成分，以主隶属度模糊离散化过程，并以次隶属度将主隶属度进一步模糊化，通过确定的主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集。
S15：对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。
具体地，离散化就是采取某种特定的方法将连续特征划分为多个子区间，并将多个子区间与候选断点关联起来。因此，对目标遥感图像的特征离散化处理可以看作是对候选断点的选择。对目标遥感图像数据进行特征离散化处理，得到最优离散化结果的过程是指通过遗传算法迭代选择候选断点；并通过各迭代过程中候选断点数目的减少幅度以及II型模糊粗糙集的平均近似精度，确定种群中个体的适应度函数；以确定的种群中个体的适应度函数评估离散化结果，并得到最优离散化结果。
在实际应用中，初始离散化方案的确定是通过获取遥感图像数据中混合像元的初始断点集确定。
本申请提供的一种基于II型模糊粗糙模型的遥感影像特征离散化方法，该方法包括：获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征；根据混合像元确定各混合像元对应各地物类型的主隶属度；根据主隶属度计算各混合像元归属于各地物类型的次隶属度；根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集；对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。 本申请实施例结合粗糙集与模糊集，以混合像元对应的主隶属度和次隶属度来描述遥感影像特征离散化过程中的模糊成分，以主隶属度模糊化离散化过程，并以次隶属度将主隶属度进一步模糊化，从而准确量化和评估混合像元的不确定性，获得更加精确地离散化结果。
本申请的一个可选实施例中，上述步骤S12中，根据混合像元确定各混合像元对应各地物类型的主隶属度，包括：
(1)迭代计算预设模糊分割矩阵的模糊均值矢量和模糊协方差矩阵，预设模糊分割矩阵由混合像元对应各地物类型的隶属度组成；
具体的，预设模糊分割矩阵可按如下公式表达：
Figure PCTCN2022105555-appb-000003
其中，Fs(Xk)表示U中第k像元Xk对s类别的隶属度，s∈{1,2,…,g}，g表示类别数目，k∈{1,2,…,n}，n表示U中像元的数目。
在实际应用中，Fs(Xk)满足：0≤Fs(Xk)≤1，
Figure PCTCN2022105555-appb-000004
q∈{1,2,…,g}。
具体地，预设模糊分割矩阵的模糊均值矢量可按如下公式表达：
Figure PCTCN2022105555-appb-000005
其中，
Figure PCTCN2022105555-appb-000006
表示模糊均值，i∈{1,2,…,m}，m表示波段个数，
Figure PCTCN2022105555-appb-000007
w表示权重，w大于等于1，xik表示第k像元在第i波段上的像元值。
具体地，预设模糊分割矩阵的模糊协方差矩阵可按如下公式表达：
Figure PCTCN2022105555-appb-000008
其中，σ mms表示模糊协方差，j∈{1,2,…,m}，
Figure PCTCN2022105555-appb-000009
具体地，迭代计算预设模糊分割矩阵的模糊均值矢量和模糊协方差矩阵包括由模糊均值矢量和模糊协方差矩阵确定预设模糊分割矩阵。
在实际应用中，预设模糊分割矩阵由像元对类别的隶属度组成，由模糊均值矢量和模糊协方差矩阵可以确定像元对类别的隶属度，像元对类别的隶属度可按如下公式表达：
Figure PCTCN2022105555-appb-000010
其中，P`(s)为第s类别出现的先验概率，
Figure PCTCN2022105555-appb-000011
在实际应用中，对于第s类别出现的先验概率的确定属于较为成熟的现有技术，本申请对此不再进行赘述。
(2)根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各混合像元对应各地物类型的主隶属度。
具体地，迭代终止条件可按如下公式表示：
max sk{|F s(X k) (θ+1)-F s(X k) (θ)|}＜ε，
其中，θ为算法的迭代步数，ε是误差阈值。
在本申请的一个可选实施例中，根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各混合像元对应各地物类型的主隶属度，包括：根据模糊分割矩阵，确定各混合像元对应的丰度，并将丰度作为各混合像元对应各地物类型的主隶属度。
具体地，各混合像元对应各地物类型的主隶属度的确定与权重及混合像元中各端元的丰度相关。
Figure PCTCN2022105555-appb-000012
其中e为自然数。对于各混合像元对应各地物类型的主隶属度可按如下公式表达：
Figure PCTCN2022105555-appb-000013
Figure PCTCN2022105555-appb-000014
其中，Ps(x)表示混合像元x对应的丰度，
Figure PCTCN2022105555-appb-000015
表示各混合像元归属于各地物类型的次隶属度，Jx表示主隶属度的取值范围，u表示一个主隶属度，
Figure PCTCN2022105555-appb-000016
表示各混合像元对应各地物类型的主隶属度的最小值，
Figure PCTCN2022105555-appb-000017
表示各混合像元对应各地物类型的主隶属度的最大值。
在实际应用中，e决定了w中权重的个数，e的值越大，Ps(x)包含的元素越多，即主隶属度的取值范围越大，与此同时，也带来了大量的计算量。权重w不仅了离散化结果的凸凹性，还控制着混合像元在各类之间的分享程度。研究结果表明，当权重为2时，聚类的效果能以较大的概率达到最好。如果e的取值大于1，则会引入不同的权重，模糊聚类的结果能够被考虑的更加全面，却无法保证引入权重带来的聚类质量，使得离散化结果存在误差并且不稳定。因此为保证离散化结果的稳定性并且降低复杂度，在一种可选实施方式中，选取e取值为1且w取值为2。此时，
Figure PCTCN2022105555-appb-000018
应该理解的是，关于w与e的取值包括但不限定于实施例中描述方式，w与e的取值只要可用于保证离散化结果的稳定性并且降低复杂度即可。
在实际应用中，对于
Figure PCTCN2022105555-appb-000019
各混合像元对应各地物类型的主隶属度可按如下公式表达：
Figure PCTCN2022105555-appb-000020
Figure PCTCN2022105555-appb-000021
在本申请的一个可选实施例中，通过迭代计算模糊均值矢量和模糊分割矩阵，确定模糊分割矩阵。通过模糊分割矩阵确定混合像元对应的丰度，并以丰度作为混合像元对应各地物类型的主隶属度。从而实现以主隶属度模糊化离散化过程。
在本申请的一个可选实施例中，在上述步骤S13中，根据主隶属度计算各混合像元归属于各地物类型的次隶属度，包括：
(1)根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定硬分割矩阵。
具体地，硬分割矩阵的确定是将满足迭代终止条件时的模糊分割矩阵中每一列的最大值修改为1，对应列中的其它值修改为0，从而完成硬分割矩阵的确定。
(2)根据硬分割矩阵，确定归属于各地物类型的像元构成的集合。
具体地，归属于各地物类型的像元构成的集合可按如下公式表达：
X s＝{x|(x∈U)∧(C s(x)＝1}，
其中，Xs表示混合像元集合中归属于第s种地物类型的像元构成的集合，Cs(x)为硬分割矩阵C中像元x对第s类别的隶属度。
在实际应用中，硬分割矩阵已确定每一列中隶属度值为1的对应位置，通过确定的每一列中隶属度值为1的对应位置，判断混合像元集合U中的混合像元是否归属于各地物类型，从而实现根据硬分割矩阵，确定归属于各地物类型的像元构成的集合。
(3)计算集合在近似空间中的上近似、下近似、正域、负域、边界域。
具体地，计算归属于各地物类型的像元构成的集合Xs在近似空间中的上近似、下近似、正域、负域、边界域包括：确定混合像元集合的等价类集合；根据确定的等价类集合计算归属于各地物类型的像元构成的集合在近似空间中的上近似、下近似、正域、负域、边界域。
具体地，混合像元集合的等价类集合可按如下公式表达：
Figure PCTCN2022105555-appb-000022
其中，U|IND(T)表示混合像元集合的等价类集合，T＝{t1,…,tm}，
Figure PCTCN2022105555-appb-000023
表示任意像元x及任意像元y属于混合像元X，
Figure PCTCN2022105555-appb-000024
表示任意波段t存在与像元x、像元y分别对应的波段等价关系。
具体地，Xs在近似空间中的上近似可按如下公式计算：
Figure PCTCN2022105555-appb-000025
其中，T*(Xs)表示Xs在近似空间中的上近似。
具体地，Xs在近似空间中的下近似可按如下公式计算：
Figure PCTCN2022105555-appb-000026
其中，T*(Xs)表示集合在近似空间中的下近似。
具体地，Xs在近似空间中的正域可按如下公式计算：
POS T(X s)＝T *(X s)，
其中，POST(Xs)表示集合在近似空间中的正域。
具体地，Xs在近似空间中的负域可按如下公式计算：
NGT T(X s)＝U-T *(X s)，
其中，NGTT(Xs)表示集合在近似空间中的负域。
具体地，Xs在近似空间中的边界域可按如下公式计算：
BN T(X s)＝T *(X s)-T *(X s)，
其中，BNT(Xs)表示集合在近似空间中的边界域。
(4)根据上近似、下近似、正域、负域、边界域确定各混合像元归属于各地物类型的次隶属度。
具体地，根据上近似、下近似、正域、负域、边界域确定各混合像元归属于各地物类型的次隶属度，包括：根据上近似、下近似、正域、负域、边界域，确定像元归属于各地物类型的像元构成的集合的概率；根据确定的像元归属于各地物类型的像元构成的集合的概率，确定各混合像元归属于各地物类型的次隶属度。
具体地，根据上近似、下近似、正域、负域、边界域，确定像元归属于各地物类型的像元构成的集合的概率是指根据上近似、下近似、正域、负域、边界域确定像元的分布情况；以确定的分布情况，根据贝叶斯定理，确定像元归属于各地物类型的像元构成的集合的概率。
示例性地，POST(Xs)是混合像元集合中归属于Xs的像元构成的集合，NGTT(Xs)是混合像元集合中不归属于Xs的像元构成的集合，BNT(Xs)是混合像元集合中不能肯定属于Xs的像元构成的集合。因此，BNT(Xs)是混合像元的不确定域。如图2所示，本申请实施例中基于II型模糊粗糙模型的遥感影像特征离散化方法的一个具体示例分析图，图中每个矩形表示U|IND(T)中的一个等价类。附图标记22所对应的圆形区域表示Xs，附图标记24所对应的八边形区域表示T*(Xs)，附图标记23所对应的矩形区域表示T*(Xs)或POST(Xs)。附图标记24所对应的八边形以外的区域表示NGTT(Xs)，附图标记24所对应的八边形以内，附图标记23所对应的矩形以外的区域表示BNT(Xs)。当像元x出现在正域或者负域时，x与Xs的关系是确定的，当x出现在边界域时，x与Xs的关系是不确定的。即，根据上近似、下近似、正域、负域、边界域确定像元的分布情况。
具体地，以确定的分布情况，根据贝叶斯定理，确定像元归属于各地物类型的像元构成的集合的概率，包括：像元归属于各地物类型的像元构成的集合的概率可按如下公式表达：
P s(x)＝P(G[s]|E[i])＝P(E[i]|G[s])P(G[s])/P(E[i])，
其中，E[i]表示像元集合上的第i个等价类，G[s]表示第s类别，P(G[s]|E[i])表示等价类i中类别s的像元所占的比例，P(E[i]|G[s])表示在所有类别为s的像元中属于等价类i的像元所占的比例，P(E[i])表示等价类i中的像元在混合像元集合U中出现的概率，P(G[s])表示类别s的像元在U中出现的概率。
具体地，根据确定的像元归属于各地物类型的像元构成的集合的概率，确定各混合像元归属于各地物类型的次隶属度，包括：各混合像元归属于各地物类型的次隶属度可按如下公 式表达：
Figure PCTCN2022105555-appb-000027
在本申请的一个可选实施例中，通过确定各地物类型的像元构成的集合，计算集合在近似空间中的上近似、下近似、正域、负域、边界域；根据上近似、下近似、正域、负域、边界域，确定像元归属于各地物类型的像元构成的集合的概率；根据确定的像元归属于各地物类型的像元构成的集合的概率，确定各混合像元归属于各地物类型的次隶属度。在这一过程中，由于离散化方案相当于把混合像元集合划分成多个等价类，处于同一个等价类的元素具有相同的属性值。则，同一个等价类内的混合像元归属于各类别的概率可以被认为是相同的，而对处于不同等价类的混合像元，它们归属于各类别的概率会存在差异，这种差异带来的不确定性是对主隶属度的进一步模糊化。因此，本申请技术方案确定次隶属度的过程，相当于通过次隶属度将主隶属度进一步模糊化，从而准确量化和评估混合像元的不确定性。
在本申请的一个可选实施例中，上述步骤S14中，对于
Figure PCTCN2022105555-appb-000028
各地物类型的II型模糊粗糙集通过如下公式表达：
Figure PCTCN2022105555-appb-000029
在本申请的一个可选实施例中，上述步骤S15中，对目标遥感图像数据进行特征离散化处理，得到最优离散化结果，包括：
(1)从遥感图像数据中获取混合像元的初始断点集；
具体地，从遥感图像数据中获取混合像元的初始断点集是指从输入的遥感影像特征中获取初始断点集。初始断点集的获取属于较为成熟的现有技术，本申请对此不再进行赘述。
(2)基于初始断点集的断点数量初始化目标遥感图像数据种群；
具体地，离散化就是采取某种特定的方法将连续特征划分为多个子区间，并将多个子区间与候选断点关联起来。因此，对目标遥感图像的特征离散化处理可以看作是对候选断点的选择，每个离散化方案对应混合像元集合上的一种划分结果。初始目标遥感图像数据种群是指以初始断点集的断点数量作为初始种群的个体长度，从而完成目标遥感图像数据种群的初始化。
示例性地，假设初始断点的数量为10，由于每个种群个体采用二进制编码，个体的长度为初始断点的数量，即每个种群个体的长度为10位，分别对应初始断点集的10个断点。
(3)对目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果；其中，与初始化后的目标遥感图像数据种群相对应的离散化方案为初始离散化方案，每个种群个体对应一个离散化结果。
具体的，与初始种群中个体相对应的离散化方案为初始离散化方案，与初始离散化方案相对应的离散化结果为初始离散化结果。
示例性地，假设目标遥感图像数据种群个体数量为50，与目标遥感图像数据种群大小相对应的离散化方案共有50个，在每次迭代中，种群中的这50个个体会经历选择、变异、交叉，即遗传算法的演化功能来更新种群，从而使得下一代的种群中的50个个体都和上一代的50个个体不相同。
在本申请的一个可选实施例中，对目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果，包括：
(1)基于混合像元间的欧氏距离确定混合像元间的模糊关系；
具体地，混合像元间的模糊关系可按如下公式表达：
Figure PCTCN2022105555-appb-000030
Figure PCTCN2022105555-appb-000031
其中，d(x,y)表示x和y的欧氏距离，xh和yh分别表示x和y在h波段上的像元值。
(2)根据模糊关系，计算II型模糊粗糙集的平均近似精度；
在本申请的一个可选实施例中，根据模糊关系，计算II型模糊粗糙集的平均近似精度，包括：根据模糊关系，计算II型模糊粗糙集的上近似与下近似；根据确定的上近似与下近似，计算II型模糊粗糙集的平均近似精度。
具体地，II型模糊粗糙集的上近似可按如下公式表达：
Figure PCTCN2022105555-appb-000032
其中，
Figure PCTCN2022105555-appb-000033
Figure PCTCN2022105555-appb-000034
u2为主隶属度中的一个，
Figure PCTCN2022105555-appb-000035
Jy2表示对应隶属度的取值范围，a(y2)表示混合像元y2的次隶属度，
Figure PCTCN2022105555-appb-000036
和
Figure PCTCN2022105555-appb-000037
分别是
Figure PCTCN2022105555-appb-000038
的主隶属度最小值和最大值。
具体地，II型模糊粗糙集的下近似可按如下公式表达：
Figure PCTCN2022105555-appb-000039
其中，
Figure PCTCN2022105555-appb-000040
Figure PCTCN2022105555-appb-000041
u1为主隶属度中的一个，
Figure PCTCN2022105555-appb-000042
Jy1表示对应隶属度的取值范围，a(y1)表示混合像元y1的次隶属度。
具体地，II型模糊粗糙集的平均近似精度可按如下公式表达：
Figure PCTCN2022105555-appb-000043
其中，
Figure PCTCN2022105555-appb-000044
为平均近似精度，
Figure PCTCN2022105555-appb-000045
表示II型模糊粗糙集的近似精度。
具体地，II型模糊粗糙集的近似精度可按如下公式表达：
Figure PCTCN2022105555-appb-000046
其中，x，y1，y2∈U。
示例性地，对于
Figure PCTCN2022105555-appb-000047
则II型模糊粗糙集的近似精度可按如下公式表达：
Figure PCTCN2022105555-appb-000048
(3)根据初始断点集的断点数量，确定与目标遥感图像数据种群个体相对应的断点数目减少的幅度；
具体的，根据初始断点集的断点数量，确定与目标遥感图像数据种群个体相对应的断点数目减少的幅度，包括：确定个体的长度为初始断点的数量；确定个体被选择的长度个数为对应个体的断点数目；确定个体的长度与个体断点数目的差值为与目标遥感图像数据种群个体相对应的断点数目减少的幅度。
示例性地，假设初始断点的数量为10，由于每个种群个体采用二进制编码，个体的长度为初始断点的数量，即每个种群个体的长度为10位，分别对应初始断点集的10个断点。二进制码中的每一位对应一个候选断点，取值‘1’和‘0’分别代表该断点被选择和未被选择。对于每个种群个体，确定二进制码被选择的个数为对应个体的断点数目，即确定值为‘1’的二进制位的个数可以得到该个体的断点数目。确定个体的长度与个体断点数目的差值为对应个体的断点数目减少的幅度，即用10减去该个体的断点数目就等于该个体断点数目减少的幅度。比如某个体的二进制编码为1110000111，与该个体相对应的断点数目为6，该个体断点数目减少的幅度为4。
具体地，断点数目减少的幅度是指各迭代过程中种群个体断点数目的减少的数量。通过比较迭代过程中种群断点数目减少的数量衡量离散效果，断点数目减少的幅度越大所对应的离散效果越好。
(4)根据断点数目减少的幅度与平均近似精度，确定II型模糊粗糙集的适应度函数；
具体地，II型模糊粗糙集的适应度函数用于计算每个目标遥感图像数据种群个体的适应度值，适应度函数由断点数目减少的幅度与II型模糊粗糙集的平均近似精度加权求和构成。
在本申请的一个可选实施例中，II型模糊粗糙集的适应度函数通过如下公式表达：
Figure PCTCN2022105555-appb-000049
其中，α和β为权重系数，|D|为以断点数目减少的幅度，
Figure PCTCN2022105555-appb-000050
为平均近似精度。
具体地，权重系数的选择是根据实际工况进行选择，一般根据数据集的特点和实验观察来判断权重设置的合理性，本申请对此不作具体限定。
(5)根据II型模糊粗糙集的适应度函数，确定II型模糊粗糙集的适应度值，并以最优适应度值对应的各目标遥感图像数据种群的个体作为所述最优离散化结果。
具体地，将多个离散化方案作为遗传算法中的种群个体，通过遗传算法的演化功能，迭代计算寻找适应度值最大的个体，并以最大的适应度值的个体作为最优适应度值的个体。最优适应度值所对应的离散化方案即为最优离散化方案。
示例性地，假设目标遥感图像数据种群中有50个个体，与种群个体相对应的适应度函数具有50个值。在每次迭代中，种群中的这50个个体会经历演化更新种群。使得下一代的种群中的50个个体都和上一代的50个个体不相同。在每次的迭代过程中，全局变量会记录50个个体中适应度值最高的个体。当下一代存在个体的适应度值高于全局变量记录的个体的适应度值时，就用具有更高适应度值的个体更新全局变量。所有的迭代经历完后，全局变量记录的就是最优的个体，与最优的个体相对应的离散化方案即为最优的离散化方案。
在本申请的一个可选实施例中，通过计算II型模糊粗糙集的平均近似精度与确定断点数目减少的幅度，从而构建II型模糊粗糙集的适应度函数。并通过遗传算法确定最优适应度值的个体，以混合像元对应的主隶属度和次隶属度来描述遥感影像特征离散化过程中的模糊成分，以主隶属度模糊化离散化过程，并以次隶属度将主隶属度进一步模糊化，并通过构建的II型模糊粗糙集的适应度函数，获得更加精确地离散化结果。
实施例2
本施例提供一种基于II型模糊粗糙模型的遥感影像特征离散化装置，如图3所示，图3是本申请一个可选实施例提供的一种基于II型模糊粗糙模型的遥感影像特征离散化装置的连接图，包括：混合像元提取单元31，主隶属度确定单元32，次隶属度确定单元33，模糊粗糙集确定单元34，最优离散化结果确定单元35。
其中，混合像元提取单元31，被配置为获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征。详细内容可参见上述任意方法实施例的步骤S11的相关描述，在此不再赘述。
主隶属度确定单元32，被配置为根据混合像元确定各混合像元对应各地物类型的主隶属度。详细内容可参见上述任意方法实施例的步骤S12的相关描述，在此不再赘述。
次隶属度确定单元33，被配置为根据主隶属度计算各混合像元归属于各地物类型的次隶属度。详细内容可参见上述任意方法实施例的步骤S13的相关描述，在此不再赘述。
模糊粗糙集确定单元34，被配置为根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集。详细内容可参见上述任意方法实施例的步骤S14的相关描述，在此不再赘述。
最优离散化结果确定单元35，被配置为对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。详细内容可参见上述任意方法实施例的步骤S15的相关描述，在此不再赘述。
本申请提供的一种基于II型模糊粗糙模型的遥感影像特征离散化装置，该装置包括：混合像元提取单元31，被配置为获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征。主隶属度确定单元32，被配置为根 据混合像元确定各混合像元对应各地物类型的主隶属度。次隶属度确定单元33，被配置为根据主隶属度计算各混合像元归属于各地物类型的次隶属度。模糊粗糙集确定单元34，被配置为根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集。最优离散化结果确定单元35，被配置为对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。本申请实施例结合粗糙集与模糊集，以混合像元对应的主隶属度和次隶属度来描述遥感影像特征离散化过程中的模糊成分，以主隶属度模糊化离散化过程，并以次隶属度将主隶属度进一步模糊化，从而准确量化和评估混合像元的不确定性，获得更加精确地离散化结果。
本申请的一个可选实施例中，主隶属度确定单元32，包括：迭代计算子单元与主隶属度确定子单元。详细内容可以参见上述任意方法实施例中关于根据混合像元确定各混合像元对应各地物类型的主隶属度的相关描述。
迭代计算子单元，被配置为迭代计算预设模糊分割矩阵的模糊均值矢量和模糊协方差矩阵，其中，预设模糊分割矩阵由混合像元对应各地物类型的隶属度组成。
主隶属度确定子单元，被配置为根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各混合像元对应各地物类型的主隶属度。
本申请的一个可选实施例中，子隶属度确定子单元，包括丰度确定子单元，被配置为根据模糊分割矩阵，确定各混合像元对应的丰度，并将丰度作为各混合像元对应各地物类型的主隶属度。详细内容可以参见上述任意方法实施例中关于根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各混合像元对应各地物类型的主隶属度的相关描述。
本申请的一个可选实施例中，次隶属度确定单元33，包括：硬分割矩阵确定子单元，像元集合确定子单元，近似空间边界确定子单元，次隶属度确定子单元。详细内容可以参见上述任意方法实施例中关于根据主隶属度计算各混合像元归属于各地物类型的次隶属度的相关描述。
硬分割矩阵确定子单元，被配置为根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定硬分割矩阵。
像元集合确定子单元，被配置为根据硬分割矩阵，确定归属于各地物类型的像元构成的集合。
近似空间边界确定子单元，被配置为计算集合在近似空间中的上近似、下近似、正域、负域、边界域。
次隶属度确定子单元，被配置为根据上近似、下近似、正域、负域、边界域确定各混合像元归属于各地物类型的次隶属度。
本申请的一个可选实施例中，最优离散化结果确定单元35，包括初始断点集获取子单元，种群初始化子单元，最优离散化结果确定子单元。详细内容可以参见上述任意方法实施例中关于对目标遥感图像数据进行离散化处理，得到最优离散化结果的相关描述。
初始断点集获取子单元，被配置为从遥感图像数据中获取混合像元的初始断点集。
种群初始化子单元，被配置为基于初始断点集的断点数量初始化目标遥感图像数据种群。其中，与初始化后的目标遥感图像数据种群相对应的离散化方案为初始离散化方案，每个种群个体对应一个离散化结果。
最优离散化结果确定子单元，被配置为对目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果。
本申请的一个可选实施例中，最优离散化结果确定子单元，包括：像元间模糊关系确定 子单元，计算子单元，断点数目减少的幅度确定子单元，适应度函数确定子单元，适应度值确定子单元。详细内容可以参见上述任意方法实施例中关于对目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果的相关描述。
像元间模糊关系确定子单元，被配置为基于混合像元间的欧氏距离确定混合像元间的模糊关系。
计算子单元，被配置为根据模糊关系，计算II型模糊粗糙集的平均近似精度。
断点数目减少的幅度确定子单元，被配置为根据初始断点集的断点数量，确定与目标遥感图像数据种群个体相对应的断点数目减少的幅度。
适应度函数确定子单元，被配置为根据断点数目减少的幅度与平均近似精度，确定II型模糊粗糙集的适应度函数。
适应度值确定子单元，被配置为根据II型模糊粗糙集的适应度函数，确定II型模糊粗糙集的适应度值，并以最优适应度值对应的各目标遥感图像数据种群的个体作为最优离散化结果。
本申请提供的一种基于II型模糊粗糙模型的遥感影像特征离散化装置，该装置包括：混合像元提取单元31，被配置为获取目标遥感图像数据，从目标遥感图像数据中提取混合像元，各混合像元分别包含多种地物类型的光谱响应特征。主隶属度确定单元32，被配置为根据混合像元确定各混合像元对应各地物类型的主隶属度。次隶属度确定单元33，被配置为根据主隶属度计算各混合像元归属于各地物类型的次隶属度。模糊粗糙集确定单元34，被配置为根据主隶属度和次隶属度，确定各地物类型的II型模糊粗糙集。最优离散化结果确定单元35，被配置为对目标遥感图像数据进行特征离散化处理，得到最优离散化结果。本申请实施例结合粗糙集与模糊集，以混合像元对应的主隶属度和次隶属度来描述遥感影像特征离散化过程中的模糊成分，以主隶属度模糊化离散化过程，并以次隶属度将主隶属度进一步模糊化，从而准确量化和评估混合像元的不确定性，获得更加精确地离散化结果。
实施例3
本申请一个实施例还提供了一种非暂态计算机存储介质，所述计算机存储介质存储有计算机可执行指令，该计算机可执行指令可执行上述任意方法实施例中所述的方法。其中，所述存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)、随机存储记忆体(Random Access Memory，RAM)、快闪存储器(Flash Memory)、硬盘(Hard Disk Drive，缩写：HDD)或固态硬盘(Solid-State Drive，SSD)等；所述存储介质还可以包括上述种类的存储器的组合。
实施例4
本申请一个实施例还提供一种计算机设备，如图4所示，图4是本申请一个可选实施例提供的一种计算机设备的结构示意图，该计算机设备可以包括至少一个处理器41、至少一个通信接口42、至少一个通信总线43和至少一个存储器44，其中，通信接口42可以包括显示屏(Display)、键盘(Keyboard)，可选通信接口42还可以包括标准的有线接口、无线接口。存储器44可以是随机存取存储器(Random Access Memory，即，易失性存储器)，也可以是非易失性存储器(non-volatile memory)，例如至少一个磁盘存储器。存储器44可选的还可以是至少一个位于远离前述处理器41的存储装置。其中处理器41可以结合图3所描述的装置，存储器44中存储应用程序，且处理器41调用存储器44中存储的程序代码，以用于执行上述任意方法实施例所述方法的步骤。
其中，通信总线43可以是外设部件互连标准(peripheral component interconnect，PCI)总线或扩展工业标准结构(extended industry standard architecture，EISA)总线等。通信总线43可以分为地址总线、数据总线、控制总线等。为便于表示，图4中仅用一条粗线表示，但并不表示仅有一根总线或一种类型的总线。
其中，存储器44可以包括易失性存储器(volatile memory)，例如随机存取存储器(random-access memory，RAM)；存储器也可以包括非易失性存储器(non-volatile memory)，例如快闪存储器(flash memory)，硬盘(hard disk drive，HDD)或固态硬盘(solid-state drive，SSD)；存储器44还可以包括上述种类的存储器的组合。
其中，处理器41可以是中央处理器(central processing unit，CPU)，网络处理器(network processor，NP)或者CPU和NP的组合。
其中，处理器41还可以进一步包括硬件芯片。上述硬件芯片可以是专用集成电路(application-specific integrated circuit，ASIC)，可编程逻辑器件(programmable logic device，PLD)或其组合。上述PLD可以是复杂可编程逻辑器件(complex programmable logic device，CPLD)，现场可编程逻辑门阵列(field-programmable gate array，FPGA)，通用阵列逻辑(generic array logic，GAL)或其任意组合。
可选地，存储器44还用于存储程序指令。处理器41可以调用程序指令，实现本申请任一实施例中所述的方法。
实施例5
本申请一个实施例还提供一种计算机程序，所述程序用于使处理器执行上述任意方法实施例所述方法的步骤。
显然，上述实施例仅仅是为清楚地说明所作的举例，而并非对实施方式的限定。对于所属领域的普通技术人员来说，在上述说明的基础上还可以做出其它不同形式的变化或变动。这里无需也无法对所有的实施方式予以穷举。而由此所引伸出的显而易见的变化或变动仍处于本发明创造的保护范围之中。

Claims (11)

  1. 一种基于II型模糊粗糙模型的遥感影像特征离散化方法，其特征在于，包括：获取目标遥感图像数据，从所述目标遥感图像数据中提取混合像元，各所述混合像元分别包含多种地物类型的光谱响应特征；根据所述混合像元确定各所述混合像元对应各地物类型的主隶属度；根据所述主隶属度计算各所述混合像元归属于各地物类型的次隶属度；根据所述主隶属度和次隶属度，确定各所述地物类型的II型模糊粗糙集；对所述目标遥感图像数据进行特征离散化处理，得到最优离散化结果。
  2. 根据权利要求1所述的基于II型模糊粗糙模型的遥感影像特征离散化方法，其特征在于，所述根据所述混合像元确定各所述混合像元对应各地物类型的主隶属度，包括：迭代计算预设模糊分割矩阵的模糊均值矢量和模糊协方差矩阵，所述预设模糊分割矩阵由混合像元对应各地物类型的隶属度组成；根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各所述混合像元对应各地物类型的主隶属度。
  3. 根据权利要求2所述的基于II型模糊粗糙模型的遥感影像特征离散化方法，其特征在于，所述根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定各所述混合像元对应各地物类型的主隶属度，包括：根据所述模糊分割矩阵，确定各所述混合像元对应的丰度，并将丰度作为各所述混合像元对应各地物类型的主隶属度。
  4. 根据权利要求1所述的基于II型模糊粗糙模型的遥感影像特征离散化方法，其特征在于，所述根据所述主隶属度计算各所述混合像元归属于各地物类型的次隶属度，包括：根据迭代计算满足迭代终止条件时的模糊分割矩阵，确定硬分割矩阵；根据所述硬分割矩阵，确定归属于各地物类型的像元构成的集合；计算所述集合在近似空间中的上近似、下近似、正域、负域、边界域；根据所述上近似、下近似、正域、负域、边界域确定各所述混合像元归属于各地物类型的次隶属度。
  5. 根据权利要求1所述的基于II型模糊粗糙模型的遥感影像特征离散化方法，其特征在于，所述对所述目标遥感图像数据进行特征离散化处理，得到最优离散化结果，包括：从所述遥感图像数据中获取所述混合像元的初始断点集；基于所述初始断点集的断点数量初始化所述目标遥感图像数据种群；对所述目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果；其中，与初始化后的所述目标遥感图像数据种群相对应的离散化方案为初始离散化方案，每个种群个体对应一个离散化结果。
  6. 根据权利要求5所述的基于II型模糊粗糙模型的遥感影像特征离散化方法，其特征在于，所述对所述目标遥感图像数据种群的个体迭代执行遗传算法，确定最优离散化结果，包括：基于所述混合像元间的欧氏距离确定混合像元间的模糊关系；根据所述模糊关系，计算所述II型模糊粗糙集的平均近似精度；根据所述初始断点集的断点数量，确定与所述目标遥感图像数据种群个体相对应的断点数目减少的幅度；根据所述断点数目减少的幅度与所述平均近似精度，确定所述II型模糊粗糙集的适应度函数；根据所述II型模糊粗糙集的适应度函数，确定所述II型模糊粗糙集的适应度值，并以最优适应度值对应的各所述目标遥感图像数据种群的个体作为所述最优离散化结果。
  7. 根据权利要求6所述的基于II型模糊粗糙模型的遥感影像特征离散化方法，其特征在于，所述II型模糊粗糙集的适应度函数通过如下公式表达：
    Figure PCTCN2022105555-appb-100001
    其中，α和β为权重系数，|D|为以断点数目减少的幅度，
    Figure PCTCN2022105555-appb-100002
    为平均近似精度。
  8. 一种基于II型模糊粗糙模型的遥感影像特征离散化装置，其特征在于，包括：混合像元提取单元，被配置为获取目标遥感图像数据，从所述目标遥感图像数据中提取混合像元，各所述混合像元分别包含多种地物类型的光谱响应特征；主隶属度确定单元，被配置为根据所述混合像元确定各所述混合像元对应各地物类型的主隶属度；次隶属度确定单元，被配置为根据所述主隶属度计算各所述混合像元归属于各地物类型的次隶属度；模糊粗糙集确定单元，被配置为根据所述主隶属度和次隶属度，确定各所述地物类型的II型模糊粗糙集；最优离散化结果确定单元，被配置为对所述目标遥感图像数据进行特征离散化处理，得到最优离散化结果。
  9. 一种非暂态计算机可读存储介质，其特征在于，所述非暂态计算机可读存储介质存储计算机指令，所述计算机指令被处理器执行时实现如权利要求1-7中任一项所述的基于II型模糊粗糙模型的遥感影像特征离散化方法。
  10. 一种计算机设备，其特征在于，包括：至少一个处理器；以及与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，从而执行如权利要求1-7中任一项所述的基于II型模糊粗糙模型的遥感影像特征离散化方法。
  11. 一种计算机程序，所述程序用于使处理器执行如权利要求1-7中任一项所述的基于II型模糊粗糙模型的遥感影像特征离散化方法。
PCT/CN2022/105555 2022-07-04 2022-07-13 基于ii型模糊粗糙模型的遥感影像特征离散化方法、装置、存储介质、计算机设备 WO2022258077A2 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210776562.XA CN114881892B (zh) 2022-07-04 2022-07-04 基于ii型模糊粗糙模型的遥感影像特征离散化方法及装置
CN202210776562.X 2022-07-04

Publications (2)

Publication Number Publication Date
WO2022258077A2 true WO2022258077A2 (zh) 2022-12-15
WO2022258077A3 WO2022258077A3 (zh) 2023-09-21

Family

ID=82683265

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/105555 WO2022258077A2 (zh) 2022-07-04 2022-07-13 基于ii型模糊粗糙模型的遥感影像特征离散化方法、装置、存储介质、计算机设备

Country Status (3)

Country Link
CN (1) CN114881892B (zh)
LU (1) LU503147B1 (zh)
WO (1) WO2022258077A2 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117111585B (zh) * 2023-09-08 2024-02-09 广东工业大学 基于容差子关系粗糙集的数控机床健康状态预测方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127784B (zh) * 2016-07-01 2018-08-07 辽宁工程技术大学 一种高分辨率遥感影像分割方法
AU2020100179A4 (en) * 2020-02-04 2020-03-19 Huang, Shuying DR Optimization Details-Based Injection Model for Remote Sensing Image Fusion
CN111428627B (zh) * 2020-03-23 2023-03-24 西北大学 一种山地地貌遥感提取方法及系统
CN112116610B (zh) * 2020-08-24 2024-02-27 中国科学院深圳先进技术研究院 一种遥感影像分割方法、系统、终端以及存储介质
CN112580483B (zh) * 2020-12-15 2023-12-01 海南大学 一种基于粗糙模糊模型的遥感影像特征离散化方法

Also Published As

Publication number Publication date
LU503147B1 (en) 2023-04-07
CN114881892A (zh) 2022-08-09
WO2022258077A3 (zh) 2023-09-21
CN114881892B (zh) 2022-10-14

Similar Documents

Publication Publication Date Title
WO2022041598A1 (zh) 一种遥感影像分割方法、系统、终端以及存储介质
WO2020062360A1 (zh) 一种图像融合分类的方法及装置
CN104239902B (zh) 基于非局部相似性和稀疏编码的高光谱图像分类方法
WO2022179083A1 (zh) 图像检测方法、装置、设备、介质和程序
CN104820841B (zh) 基于低阶互信息和光谱上下文波段选择的高光谱分类方法
EP4425376A1 (en) Method and apparatus for searching for neural network ensemble model, and electronic device
CN111340135B (zh) 基于随机投影的肾小肿块分类方法
CN114582470A (zh) 一种模型的训练方法、训练装置及医学影像报告标注方法
JP2015203946A (ja) ヒストグラムの重心を計算する方法
EP4394724A1 (en) Image encoder training method and apparatus, device, and medium
WO2022258077A2 (zh) 基于ii型模糊粗糙模型的遥感影像特征离散化方法、装置、存储介质、计算机设备
CN104077765B (zh) 图像分割装置、图像分割方法
Choy et al. Fuzzy bit-plane-dependence image segmentation
CN117593253A (zh) 乳腺病理图像有丝分裂检测方法、系统、存储介质、设备
CN111694954A (zh) 图像分类方法、装置和电子设备
CN105654126A (zh) 一种计算设备、核矩阵评估方法以及多核学习方法
CN105844299B (zh) 一种基于词袋模型的图像分类方法
Zhang et al. Multi‐feature fusion of deep networks for mitosis segmentation in histological images
CN110837853A (zh) 一种快速分类模型构建方法
CN107492101B (zh) 基于自适应构造最优图的多模态鼻咽肿瘤分割算法
CN115691676A (zh) 一种分析组织细胞成分的方法、装置及存储介质
CN113177602B (zh) 图像分类方法、装置、电子设备和存储介质
US20150186797A1 (en) Data reduction in nearest neighbor classification
CN112037167B (zh) 一种基于影像组学和遗传算法的目标区域确定系统
JP2016062249A (ja) 識別辞書学習システム、認識辞書学習方法および認識辞書学習プログラム