CN102073882A - Method for matching and classifying spectrums of hyperspectral remote sensing image by DNA computing - Google Patents
Method for matching and classifying spectrums of hyperspectral remote sensing image by DNA computing Download PDFInfo
- Publication number
- CN102073882A CN102073882A CN 201110028401 CN201110028401A CN102073882A CN 102073882 A CN102073882 A CN 102073882A CN 201110028401 CN201110028401 CN 201110028401 CN 201110028401 A CN201110028401 A CN 201110028401A CN 102073882 A CN102073882 A CN 102073882A
- Authority
- CN
- China
- Prior art keywords
- dna
- spectrum
- training sample
- spectral
- target
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Abstract
The invention provides a method for matching and classifying spectrums of a hyperspectral remote sensing image by DNA computing. The DNA computing idea is applied to the spectral encoding matching algorithm, the existing ground object spectrum data are converted into a corresponding DNA chain parameter according to the principle of optimality, and a molecular genetic information model which is based on the DNA encoding mechanism and the DNA controlling mechanism is established, thus the spectrums of the hyperspectral remote sensing data are matched and classified by adopting the DNA computing. The method uses the high-dimension characteristic of the hyperspectral remote sensing data to match and classify the spectrums, solves the problems caused by large data amount and high data dimension during the hyperspectral remote sensing image processing and fully exerts the capacity of fine distinguishing the variety of the ground object of the hyperspectral data in the spectrum field. The method combines and optimizes the characteristic coding by the DNA-gene based control technology, can accommodate the spectrum diversity and the spectrum curve error and can be used for matching and classifying the spectrums in an intelligent, rapid and self-adaptive mode.
Description
Technical field
The invention belongs to the remote sensing image processing technology field, the DNA that particularly relates to a kind of target in hyperspectral remotely sensed image calculates the Spectral matching sorting technique.
Background technology
High-spectrum remote sensing data has comprised the spectral signature information of enriching, and theoretically, utilizes the accurate classification of pixel in the process decision chart picture that high-spectral data can be clear and definite, the uncertainty when reducing terrain classification identification.Traditional mode identification method, be with the spatial relationship of pixel or in feature space the distribution tissue typing algorithm of each classification pixel, such sorting algorithm shows inadaptability for high-dimensional high-spectral data.In order to utilize abundant spectral signature to finish high spectral information decipher, the Spectral matching technology is invented.The Spectral matching technology is the sorting algorithm towards the high spectrum image characteristics, utilizes spectroscopic data known or that gather in the library of spectra on the spot, by extracting the spectral signature of the curve of spectrum, adopts Matching Algorithm to come ground cover type in the recognition image.Finishing Spectral matching need solve two problems: the first, and there is the diversity phenomenon in the curve of spectrum, selects the optimum matching sample.Under different conditions, though curve of spectrum overall trend is consistent, curve of spectrum details there are differences with a kind of material.If selected inappropriate matched sample, this sample does not possess the significant characteristic features of this classification object spectrum, and the result that the Spectral matching algorithm obtains will occur than mistake; Secondly, extract curve of spectrum characteristic feature information, get rid of the unstable feature of the curve of spectrum.The Spectral matching feature extracting method of Spectral matching is the basis of the distance measure of Spectral matching, so curve of spectrum feature extraction need take into account the overall trend and the gradient feature of spectrum.
Traditional Spectral matching method comprises spectrum angle coupling, spectral absorption characteristics coupling and optical spectrum encoded coupling.Spectrum angle matching process is the generalized angle of atural object vector more to be identified and known atural object vector, determines the ownership of every class atural object; The spectral absorption characteristics coupling is that the whole of sample spectrum or its certain part are carried out the fundamental function match of the curve of spectrum, calculates the probability that pixel spectrum is under the jurisdiction of a certain sample by the degree of fitting that calculates between pixel spectrum and the sample light spectrum signature function; Optical spectrum encoded coupling is by the given Threshold Segmentation curve of spectrum, and spectral signature is converted into the form of binary-coding, determines the spectral class method for distinguishing by hamming (Hamming) distance of calculating between optical spectrum encoded.What the present invention studied is the method for optical spectrum encoded coupling.Traditional optical spectrum encoded coupling refers generally to the binary-coding coupling, and this method is insufficient owing to spectral signature is extracted, and only can simply discern the classification of spectrum, can't realize high-precision Spectral matching and image classification.
Some improvement technology have appearred in this area at present:
In order to improve the optical spectrum encoded quantity of information that comprises original spectrum, optical spectrum encoded (SDFC) method of a kind of two-value based on spectral analysis optical spectrum encoded (SPAM) method, a kind of coding based on spectral signature (SFBC) method and a kind of spectral differences branch feature has been proposed.These methods have increased the coding bit quantity to a wave band, taken into account the correlativity between the spectral band neighborhood simultaneously, improved the precision of Spectral matching to a certain extent, but because optical spectrum encoded coupling, can't utilize the relevant information between statistical models and neighbor, if the difference of object spectrum curve of the same race surpasses certain scope, the precision of these optical spectrum encoded matching process will be difficult to satisfy the Classification and Identification demands of applications.
It is a kind of novel Intelligentized method that DNA calculates, and its initiative thought is 1994 (1994) that proposed on Science by the Adleman professor of American South University of California, thus the new era of having started dna computer.DNA calculates and utilizes dna encoding to represent complicated knowledge or system, have self-generating, the self organizing function of analyzing and imitate the hereditary information regulator control system further, in evolution, can obtain and refresh one's knowledge, used (Lipton, 1995 widely in engineering fields such as pattern-recognition, fuzzy control, decision problems; Faulhammer etc., 2000; Ren etc., 2001; Chen etc., 2003; Benenson etc., 2004).DNA calculates with the high-spectrum remote-sensing Spectral matching and has inner link in itself, therefore DNA can be calculated the high-spectrum remote-sensing Spectral matching field of introducing.Yet in target in hyperspectral remotely sensed image Spectral matching field, DNA calculates and is not also well used, and does not therefore also have the DNA of real meaning to calculate the appearance of Spectral matching method.
Summary of the invention
At the problem that above-mentioned prior art exists, the DNA that the present invention proposes a kind of target in hyperspectral remotely sensed image calculates the Spectral matching sorting technique.
The DNA of target in hyperspectral remotely sensed image that technical scheme of the present invention provides calculates the Spectral matching sorting technique, comprises the steps:
Step 1 is selected the required target in hyperspectral remotely sensed image that exercises supervision and classify;
Step 2 is for the selected target in hyperspectral remotely sensed image of step 1 makes up the training sample library of spectra;
Step 3 is carried out dna encoding to all pixel spectrum of the selected target in hyperspectral remotely sensed image of step 1, obtains Hyperspectral imaging DNA message sense; Training sample spectrum in the step 2 gained training sample library of spectra is carried out dna encoding, obtain training spectrum DNA message sense;
Step 4, adopt the random dna code word combination, from training spectrum DNA message sense, produce many group DNA populations, every group of DNA population comprises the DNA individuality of similar number, each DNA individuality comprises the DNA message sense that equates with the class categories number, and these DNA populations constitute set as initial DNA database;
Step 5, calculate the distance between DNA message sense in the initial DNA database and the training spectrum DNA message sense, according to the minor increment training spectrum DNA message sense of presorting, then according to the classification information evaluation of the training sample spectrum precision of presorting, calculate the kappa coefficient of each DNA individuality in the initial DNA database, kappa coefficient maximal value is an optimized individual in every group of DNA population, preserves to organize respectively in the DNA population that best DNA is individual to be best DNA database;
Step 6 judges whether to satisfy the feature selecting end condition, confirms that then step 5 obtains best DNA database if satisfy, and writes down all DNA message senses in the best DNA database, enters step 8 then, does not then enter step 7 if do not satisfy;
Step 7 obtains the DNA population set of variation back by initial DNA database being carried out the DNA genetic manipulation, is new initial DNA database with the DNA cluster cooperation of variation back, returns step 5 then;
Step 8, the distance between DNA message sense and the Hyperspectral imaging DNA message sense in the calculating optimum DNA database according to pixels, and, obtain the supervised classification file with each pixel that minor increment corresponding class information is given target in hyperspectral remotely sensed image.
And in the step 3, dna encoding adopts the mode of spectral absorption characteristics coding and the combination of spectrum gradient feature coding to realize, the result who is connected in series spectrum gradient feature coding by the result with the spectral absorption characteristics coding obtains corresponding DNA message sense,
Spectral absorption characteristics coding formula is as follows:
Wherein,
Represent pixel spectrum vector to average;
The pixel property value is divided into
,
The pixel in two intervals is determined in two intervals, respectively the spectral value in two intervals is got average again, obtains two new threshold values
With
, finally form four intervals
,
,
,
,
,
Represent respectively
iBand spectrum value, curve of spectrum minimum value and curve of spectrum maximal value,
Spectrum gradient feature coding formula is as follows:
Wherein as follows for the mathematical description of type:
Qi Zhong ⊿ represents spectrum deviation tolerance value,
,
,
Represent respectively
iBand spectrum value,
I-1Band spectrum value and
I+1The band spectrum value.
And, in the step 2, training sample spectrum acquisition mode in the training sample library of spectra is, high-definition remote sensing image data according to the target in hyperspectral remotely sensed image correspondence, determine the position distribution of training sample reality and required class categories number, from target in hyperspectral remotely sensed image, choose the respective classes curve of spectrum, as training sample spectrum.
And in the step 2, the training sample spectrum acquisition mode in the training sample library of spectra is to investigate on the spot according to target in hyperspectral remotely sensed image, with gathering the high-spectral data of gained on the spot, as training sample spectrum.
The present invention introduces in the optical spectrum encoded matching algorithm by the thought that DNA is calculated, is corresponding D NA chain parameter according to the principle of optimality with existing object spectrum data conversion, be based upon the hereditary information model on the molecular level, realize the high-spectrum remote sensing data Spectral matching classification of calculating based on DNA based on dna encoding mechanism and DNA control mechanism.This method utilizes the high-dimensional feature of high-spectrum remote sensing data to carry out match classifying, both solved in the high-spectrum remote sensing processing procedure because big, the high problem of bringing of data dimension of its data volume has fully played high-spectral data again in the meticulous ability of distinguishing the ground species of spectral domain; The curve of spectrum that this method utilizes the dna encoding technology to extract than horn of plenty significantly absorbs feature and gradient characteristic information, and utilization is carried out Combinatorial Optimization based on DNA genetically manipulated technology to feature coding, spectrum diversity and curve of spectrum error be can contain to a certain extent, spectrum intellectuality, quick, Adaptive matching assorting process realized.
Description of drawings
Fig. 1 is the process flow diagram of the embodiment of the invention.
Fig. 2 is a DNA mutation operation synoptic diagram.
Fig. 3 is a DNA interlace operation synoptic diagram.
Fig. 4 is dna structure mutation operation deletion operator synoptic diagram.
Fig. 5 inserts the operator synoptic diagram for the dna structure mutation operation.
Fig. 6 is inverted the operator synoptic diagram for the dna structure mutation operation.
Fig. 7 is that the DNA of high-spectrum remote sensing data calculates Spectral matching principle of classification synoptic diagram.
Fig. 8 is the dna encoding synoptic diagram of the embodiment of the invention.
Fig. 9 is two class curve of spectrum dna encoding comparison diagrams, and wherein Fig. 9 a represents that spectrum 1 absorbs the characteristic DNA coding, and Fig. 9 b represents that spectrum 2 absorbs the characteristic DNA coding, and Fig. 9 c represents spectrum 1 gradient characteristic DNA coding, and Fig. 9 d represents spectrum 2 gradient characteristic DNAs coding.
Embodiment
The technical scheme that the present invention proposes is at first to utilize dna encoding to extract the entrained terrestrial object information of spectrum, obtain the exemplary spectrum coding of all kinds of atural objects by DNA operation training from sample spectrum then, set up atural object dna encoding storehouse, calculate the fuzzy rule that principle makes up by DNA, realize the DNA Spectral matching Classification and Identification of all kinds of atural objects.For the ease of understanding the present invention, at first provide theoretical foundation of the present invention:
At occurring in nature, organism surface reveals the diversity of different plant species and the similarity of same species, and this all is that (Deoxyribonucleic acid DNA) determines by the inhereditary material DNA (deoxyribonucleic acid) in the biosome.4 kinds of bases are arranged among the DNA, promptly adenine (adenine, A), guanine (guanine, G), cytimidine (cytosine, C) and thymine (thymine, T), the various combination between each base has just constituted unusual abundant information.DNA comprises a large amount of genetic codes, conveys hereditary information by biochemical reaction, and this process is one of essential characteristic of biological phenomena.Biosome is realized the transmission and the expression of hereditary information by the dna sequence dna shirtsleeve operation.The structure of the complexity that biosome had is actually the raw information that is coded in the dna sequence dna and obtains through some simple handle.And, ask a value that contains the calculable functions of variable also can realize by asking a series of the compound of simple function that contain variable in art of mathematics.This is an important general character of biological intelligence and mathematical procedure, also is the starting point that DNA calculates.Therefore, the essence that DNA calculates is exactly to utilize a large amount of different making nucleic acid molecular hybridizations, produces the result of a kind of combination of similar certain mathematics manipulation, and to its process of screening.Based on the biological DNA theory of computation, a kind of artificial DNA computing method are invented.This method mainly comprises two parts: DNA information coding and the operation of DNA gene genetic.DNA information coding comprises an information extraction process, and the operation of DNA gene genetic comprises an information matches optimizing process.According to this theory, the sorting technique of the present invention design is referring to Fig. 7: by the curve of spectrum is carried out dna encoding, the DNA chromosome that obtains representing by A, G, C, coded strings that T constitutes (as shown in FIG. " GAT GCC TGT TGC CGA TCG TTC CGC TAA AT ... "), some information can significantly be distinguished the DNA classification information in the DNA chromosome, is called gene (GENE).Operate by DNA, the recombinant DNA message sense, and conspicuousness information transcribed on the typical information chain, by decipher DNA message sense, absorption feature after being optimized, by absorbing the distance measure (promptly calculating fitness) between the feature between comparison dna database and the pixel dna encoding, finally obtain the high-spectrum remote sensing classification results.
Below in conjunction with accompanying drawing and embodiments of the invention, technical solution of the present invention is elaborated.Referring to Fig. 1, the implementation procedure of embodiment is as follows: step 1, at first select the required target in hyperspectral remotely sensed image that exercises supervision and classify.During concrete enforcement, Software tool exploitation remote sensing image handling procedures such as suggestion employing visual c++6.0 are realized technical scheme provided by the invention.Can set, after ejecting the image parameters dialog box, by input image width, highly, wave band number and data type open the target in hyperspectral remotely sensed image of the required classification that exercises supervision, realize selecting.This opening procedure belongs to the image input process, belongs to the active computer software engineering.Total wave band number of target in hyperspectral remotely sensed image adopts among the embodiment
Mark.
Step 2 is for the selected target in hyperspectral remotely sensed image of step 1 makes up the training sample library of spectra.
The source of training sample has two kinds, and a kind of is the high resolution image data corresponding with target in hyperspectral remotely sensed image, via satellite or aviation take and to obtain.During specific implementation,, determine the position distribution of training sample reality and required class categories number, from target in hyperspectral remotely sensed image, choose the respective classes curve of spectrum, as training sample spectrum according to the high-definition remote sensing image data of target in hyperspectral remotely sensed image correspondence.Another kind is to investigate on the spot, can use the spectra collection instrument to gather spectrum, more convenient processing to target in hyperspectral remotely sensed image on the spot.During specific implementation, in the step 2, the training sample spectrum acquisition mode in the training sample library of spectra is to investigate on the spot according to target in hyperspectral remotely sensed image, with gathering the high-spectral data of gained on the spot, as training sample spectrum.Certainly, when investigating, also can determine the class categories number on the spot.During concrete enforcement, adopting wherein a kind of source separately or in conjunction with two kinds of sources, make up the training sample library of spectra, all is feasible.
Embodiment opens up a sample array SampleArray in computing machine, deposit the training sample spectroscopic data in the sample array.This array type is structure ROI(sample interested district) type, structure comprises sample data and two structure variablees of sample data classification, the sample data class variable is used to preserve the sample class number.During concrete enforcement, can preestablish some algorithm parameters, mainly comprise: the chromosome number n in each population, maximum iteration time nIteration and optimal adaptation degree threshold value Threshold by human-computer interaction interface.Set up the executive routine that activates this algorithm behind the algorithm parameter.
Step 3 is utilized the spectral signature dna encoding, with high spectrum pixel spectrum and the optical spectrum encoded DNA message sense that becomes of training sample.
Referring to Fig. 8, dna encoding adopts the mode of spectral absorption characteristics coding and the combination of spectrum gradient feature coding to realize among the embodiment, and the result who is connected in series spectrum gradient feature coding by the result with the spectral absorption characteristics coding obtains corresponding DNA message sense.The curve of spectrum is designated as array DNA_A by the result of spectral absorption characteristics coding, and code length is L; The curve of spectrum is designated as array DNA_A by the result of spectrum gradient feature coding, and length is L-2, and the dna encoding entire length is 2L-2.Wherein, L equals total wave band number of target in hyperspectral remotely sensed image
No matter from the pixel spectrum of target in hyperspectral remotely sensed image, or the training sample spectrum that provides of training sample library of spectra, be consistent to the processing mode of the curve of spectrum.When for example all pixel spectrum of the selected target in hyperspectral remotely sensed image of step 1 being carried out dna encoding, adopt the mode of spectral absorption characteristics coding and the combination of spectrum gradient feature coding, be connected in series the result who this pixel spectrum is carried out spectrum gradient feature coding by the result who certain pixel spectrum of target in hyperspectral remotely sensed image is carried out the spectral absorption characteristics coding, obtain the DNA message sense of this pixel.Curve of spectrum horizontal ordinate among Fig. 9 is wavelength (nm of unit), and ordinate is reflectivity (%), is modal spectrum expression way.During concrete enforcement, can also adopt other dna encoding modes, for example only use the spectral absorption characteristics dna encoding or only use spectrum gradient characteristic DNA coding.
Embodiment opens up array CodePix in computing machine, the dna encoding data after the storage target in hyperspectral remotely sensed image coding; Open up array CodeData, the data of storage training sample spectrum after dna encoding.DNA absorbs 3 threshold values of feature coding and passes through spectrum overall trend adaptive setting:
Represent pixel spectrum vector to average;
The pixel property value is divided into
,
The pixel in two intervals is determined in two intervals, respectively the spectral value in two intervals is got average again, obtains two new threshold values
With
, finally form four intervals
,
,
,
The gradient deviation Rong Xian Zhi ⊿ of dna ladder degree feature coding can reduce noise to optical spectrum encoded interference, makes coding extract the minutia information of spectrum essence.The spectrum deviation Rong Xian Zhi ⊿ that sets among the embodiment is determined by following formula:
Wherein,
,
Represent respectively
iBand spectrum value and
I-1The band spectrum value,
It is total wave band number of target in hyperspectral remotely sensed image.
During concrete enforcement, can also adopt other modes to calculate spectrum deviation tolerance value, for example in above-mentioned formula, add a weight beta, so that self-adaptation is adjusted threshold value.The suggestion value of β is [1,2].
Spectral absorption characteristics coding formula is as follows:
Wherein,
Represent pixel spectrum vector to average;
The pixel property value is divided into
,
The pixel in two intervals is determined in two intervals, respectively the spectral value in two intervals is got average again, obtains two new threshold values
With
, finally form four intervals
,
,
,
,
,
Expression respectively,
Spectrum gradient feature coding formula is as follows:
Wherein as follows for the mathematical description of type:
For this coded system effect is described, how the present invention illustrates according to the meticulous identification curve of spectrum of coding result with reference to Figure 10.Spectrum 1 is similar on the spectrum overall trend to spectrum 2, but there are differences on spectral details.Fig. 9 a represents that spectrum 1 absorbs characteristic DNA and is encoded to AGGGG, and Fig. 9 b represents that spectrum 2 absorbs characteristic DNA and is encoded to AGGGG, and spectrum 1 is identical on the spectral absorption characteristics coding with 2; Fig. 9 c represents that spectrum 1 gradient characteristic DNA is encoded to AAT, and Fig. 9 d represents that spectrum 2 gradient characteristic DNAs are encoded to AGG, spectrum 1 and 2 variant on spectrum gradient characteristic DNA coding.B1, b2, b3 represent wave band 1, wave band 2 and wave band 3 respectively among Fig. 9,
,
With
, represent 3 threshold values of spectral absorption characteristics dna encoding respectively.
Step 4, adopt the random dna code word combination, from training spectrum DNA message sense, produce many group DNA populations, every group of DNA population comprises the DNA individuality of similar number, each DNA individuality comprises the DNA message sense that equates with the class categories number, and these DNA populations constitute set as initial DNA database.In this step, the present invention has finished the initialization population.
Embodiment is N with the group number scale of DNA population, and wherein N value span is generally [30,100], and every group of DNA population comprises n individuality, comprises the DNA message sense that equates with Hyperspectral imaging class categories number known to the step 2 in the DNA individuality that each produces at random; The many groups DNA set that population constitutes that produces as initial DNA database, is stored among the array DNALib.Every group of DNA population comprises individual number n, and promptly the chromosome number in said each population of preamble advises that value is 20~30.
Step 5, calculate the distance between DNA message sense in the initial DNA database and the training spectrum DNA message sense, according to the minor increment training spectrum DNA message sense of presorting, then according to the classification information evaluation of the training sample spectrum precision of presorting, calculate the kappa coefficient of each DNA individuality in the initial DNA database, kappa coefficient maximal value is an optimized individual in every group of DNA population, preserves to organize respectively in the DNA population that best DNA is individual to be best DNA database.This step is finished the task that DNA translated and calculated fitness.
Training sample has been known classification information as Given information when step 2 is collected training sample spectrum, therefore can be used for estimating the precision of presorting.Among the embodiment, distance measure adopts the kappa coefficient of initial DNA database D NALib classification based training spectrum DNA message sense CodeData, promptly asks for fitness.Kappa coefficient maximal value is an optimized individual in every group of DNA population, preserves and respectively organizes the individual best DNA database array Best_DNALib of arriving of best DNA in the DNA population.The kappa coefficient is a final index that proposes on two parameters of user's precision and cartographic accuracy combining, and its implication is exactly the precision problem that is used for estimating classification image.Concrete calculating belongs to prior art, and the present invention will not give unnecessary details.
Step 6 judges whether to satisfy the feature selecting end condition, confirms that then step 5 obtains best DNA database if satisfy, and writes down all DNA message senses in the best DNA database, enters step 8 then, does not then enter step 7 if do not satisfy.
Embodiment is by judging whether to reach maximum iteration time nIteration, and perhaps the fitness of best DNA database array Best_DNALib realizes judging whether to satisfy the feature selecting end condition greater than optimal adaptation degree threshold value Threshold.Satisfy the feature selecting end condition and then deposit DNA message sense coding in the best DNA database in array Best_DNALib.
Step 7 obtains the DNA population set of variation back by initial DNA database being carried out the DNA genetic manipulation, is new initial DNA database with the DNA cluster cooperation of variation back, returns step 5 then.
Embodiment enters step 7 in the judged result of step 6 when not satisfying, and then selects by DNA, intersects, and DNA operation operators such as variation and structure variation obtain variation back DNA population and gather; The back DNA population set that will make a variation is updated to initial DNA database D NALib, returns then and continues execution in step five, up to satisfy the feature selecting end condition in step 6, determines to obtain to enter step 8 behind the best DNA database.
Step 8, the distance between DNA message sense and the Hyperspectral imaging DNA message sense in the calculating optimum DNA database according to pixels, and, obtain the supervised classification file with each pixel that minor increment corresponding class information is given target in hyperspectral remotely sensed image.
Hyperspectral imaging DNA message sense comes from the processing of step 2 to all pixel spectrum of target in hyperspectral remotely sensed image, the pixel of therefore pressing target in hyperspectral remotely sensed image is computed range one by one, can obtain the classification of each pixel, thereby finally obtain classification chart target in hyperspectral remotely sensed image.
For the sake of ease of implementation, the invention provides step 7 and obtain the specific implementation of variation back DNA population set, the selection operation that adopts is the roulette method: i.e. the probable value of giving corresponding ratio according to the size of the fitness function value of each DNA population individuality, these probability distribution are [0,1] in the domain value range, selection operation produces [0, a 1] uniform random number at random each time, and this random number is determined selected operation operator as select finger.Operation operator can comprise mutation operator, crossover operator, structure variation operator, and can designing more when specifically implementing, the multioperation operator carries out different DNA operations.
Mutation operator is the random variation of gene unit, produces the variation position at random from DNA population individuality, selects the variation mode then on appointed positions at random, finishes mutation operation.For example, there is 3 types variation mode: A-in A〉G, A-〉C, A-〉T.As Fig. 2, AGTATCCGATGCCGC is AGTATCGGATGCCGC through variation.
Crossover operator is an intersection condition code cross method, generates the intersection condition code earlier at random, is template enforcement interlace operation with the intersection condition code then.Can open up the array Crossover_array with the dna encoding equal in length, give each assignment 0 or 1 of this array at random, produce and intersect the condition code array.Corresponding intersection condition code is 1 on the dna encoding position if two are waited to intersect, and then two DNA message senses exchange in this locational value; If intersection condition code corresponding on the coding site is 0, then interlace operation can not take place in this position.As shown in Figure 3, wait to intersect DNA message sense TGAGGCCGTAGTACGATACGTAGAT and AGTATGAACTGCACGCCGTACTACT, through intersecting condition code 0001110011001110001100011, the result that obtains intersecting is: TGAATGCGCTGTACGATATATAGCT and AGTGGCAATAGCACGCCGCGCTAAT.
The structure variation operator has 3 kinds of forms: deletion, insertion and inversion, these 3 kinds of operations are to carry out on selected at random genetic fragment basis.
Deletion action is meant selects the information segment of designated length at random in the DNA message sense, it is deleted from the DNA chain, and replenish the dna encoding for respective length at random at DNA message sense end.As Fig. 4, for message sense TGAGGCCGATGTACG designated length be 2 wait to delete fragment GA, carry out deletion through the deletion operator, and after deletion action the message sense afterbody to replenish length be 2 random dna Segment A C, obtain TGAGGCCTGTACGAC.
Insert operation and be meant, generate the DNA information segment of designated length at random, produce the insertion position at random, the DNA information segment is inserted in the DNA message sense, then with DNA message sense afterbody deletion respective length dna fragmentation according to the insertion position.As Fig. 5, for DNA message sense TGAGGCCGATGTACG, specifying the insertion point and inserting fragment is ATC, carries out through the insertion operator and inserts, and delete unnecessary Segment A CG, obtains TGAGGCCATCGATGT.
Be inverted operation and be meant, from the DNA message sense, select the DNA information segment of designated length at random, will implant in this DNA message sense after this information segment reversing.As Fig. 6, for DAN message sense TGAGGCCGATGTACG, Xuan Ding inversion fragment is GCCGA at random, and the result who obtains through the inversion operator is TGAGAGCCGTGTACG.
By selection, variation, intersection and structure variation operator, make DNA calculate the optimized search process that is implemented in the dna encoding space.
Claims (4)
1. the DNA of a target in hyperspectral remotely sensed image calculates the Spectral matching sorting technique, it is characterized in that: may further comprise the steps,
Step 1 is selected the required target in hyperspectral remotely sensed image that exercises supervision and classify;
Step 2 is for the selected target in hyperspectral remotely sensed image of step 1 makes up the training sample library of spectra;
Step 3 is carried out dna encoding to all pixel spectrum of the selected target in hyperspectral remotely sensed image of step 1, obtains Hyperspectral imaging DNA message sense; Training sample spectrum in the step 2 gained training sample library of spectra is carried out dna encoding, obtain training spectrum DNA message sense;
Step 4, adopt the random dna code word combination, from training spectrum DNA message sense, produce many group DNA populations, every group of DNA population comprises the DNA individuality of similar number, each DNA individuality comprises the DNA message sense that equates with the class categories number, and these DNA populations constitute set as initial DNA database;
Step 5, calculate the distance between DNA message sense in the initial DNA database and the training spectrum DNA message sense, according to the minor increment training spectrum DNA message sense of presorting, then according to the classification information evaluation of the training sample spectrum precision of presorting, calculate the kappa coefficient of each DNA individuality in the initial DNA database, kappa coefficient maximal value is an optimized individual in every group of DNA population, preserves to organize respectively in the DNA population that best DNA is individual to be best DNA database;
Step 6 judges whether to satisfy the feature selecting end condition, confirms that then step 5 obtains best DNA database if satisfy, and writes down all DNA message senses in the best DNA database, enters step 8 then, does not then enter step 7 if do not satisfy;
Step 7 obtains the DNA population set of variation back by initial DNA database being carried out the DNA genetic manipulation, is new initial DNA database with the DNA cluster cooperation of variation back, returns step 5 then;
Step 8, the distance between DNA message sense and the Hyperspectral imaging DNA message sense in the calculating optimum DNA database according to pixels, and, obtain the supervised classification file with each pixel that minor increment corresponding class information is given target in hyperspectral remotely sensed image.
2. DNA according to claim 1 calculates the Spectral matching sorting technique, it is characterized in that: in the step 3, dna encoding adopts the mode of spectral absorption characteristics coding and the combination of spectrum gradient feature coding to realize, the result who is connected in series spectrum gradient feature coding by the result with the spectral absorption characteristics coding obtains corresponding DNA message sense
Spectral absorption characteristics coding formula is as follows:
Wherein,
Represent pixel spectrum vector to average;
The pixel property value is divided into
,
The pixel in two intervals is determined in two intervals, respectively the spectral value in two intervals is got average again, obtains two new threshold values
With
, finally form four intervals
,
,
,
,
,
Represent respectively
iBand spectrum value, curve of spectrum minimum value and curve of spectrum maximal value,
Spectrum gradient feature coding formula is as follows:
Wherein as follows for the mathematical description of type:
3. DNA according to claim 1 and 2 calculates the Spectral matching sorting technique, it is characterized in that: in the step 2, training sample spectrum acquisition mode in the training sample library of spectra is, high-definition remote sensing image data according to the target in hyperspectral remotely sensed image correspondence, determine the position distribution of training sample reality and required class categories number, from target in hyperspectral remotely sensed image, choose the respective classes curve of spectrum, as training sample spectrum.
4. DNA according to claim 1 and 2 calculates the Spectral matching sorting technique, it is characterized in that: in the step 2, training sample spectrum acquisition mode in the training sample library of spectra is, investigate on the spot according to target in hyperspectral remotely sensed image, with gathering the high-spectral data of gained on the spot, as training sample spectrum.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110028401 CN102073882A (en) | 2011-01-27 | 2011-01-27 | Method for matching and classifying spectrums of hyperspectral remote sensing image by DNA computing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110028401 CN102073882A (en) | 2011-01-27 | 2011-01-27 | Method for matching and classifying spectrums of hyperspectral remote sensing image by DNA computing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102073882A true CN102073882A (en) | 2011-05-25 |
Family
ID=44032416
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 201110028401 Pending CN102073882A (en) | 2011-01-27 | 2011-01-27 | Method for matching and classifying spectrums of hyperspectral remote sensing image by DNA computing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102073882A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102445712A (en) * | 2011-11-22 | 2012-05-09 | 成都理工大学 | Character window weighting related spectrum matching method facing rocks and minerals |
CN102663752A (en) * | 2012-04-11 | 2012-09-12 | 南京理工大学 | SAM weighted KEST hyperspectral anomaly detection algorithm |
CN104036293A (en) * | 2014-06-13 | 2014-09-10 | 武汉大学 | Rapid binary encoding based high resolution remote sensing image scene classification method |
CN104331482A (en) * | 2014-11-07 | 2015-02-04 | 武汉大学 | Method and system for hyper-spectrum matching based on feature index |
CN104867154A (en) * | 2015-05-29 | 2015-08-26 | 华中科技大学 | Remote-sensing image quality evaluation method based on gradient |
CN107798286A (en) * | 2017-07-13 | 2018-03-13 | 西安电子科技大学 | High spectrum image evolution sorting technique based on marker samples position |
CN111062888A (en) * | 2019-12-16 | 2020-04-24 | 武汉大学 | Hyperspectral image denoising method based on multi-target low-rank sparsity and spatial-spectral total variation |
CN112434685A (en) * | 2021-01-28 | 2021-03-02 | 中国科学院空天信息创新研究院 | Remote sensing image information fine extraction method based on earth surface elements |
CN112598265A (en) * | 2020-12-18 | 2021-04-02 | 武汉大学 | Decoupling risk estimation-based rapid detection method for hyperspectral pine nematode disease of unmanned aerial vehicle |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1873661A (en) * | 2006-06-29 | 2006-12-06 | 武汉大学 | Not supervised classification process of artificial immunity in remote sensing images |
CN101339619A (en) * | 2008-08-11 | 2009-01-07 | 重庆大学 | Dynamic feature selection method for mode classification |
-
2011
- 2011-01-27 CN CN 201110028401 patent/CN102073882A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1873661A (en) * | 2006-06-29 | 2006-12-06 | 武汉大学 | Not supervised classification process of artificial immunity in remote sensing images |
CN101339619A (en) * | 2008-08-11 | 2009-01-07 | 重庆大学 | Dynamic feature selection method for mode classification |
Non-Patent Citations (1)
Title |
---|
《遥感学报》 20101031 焦洪赞 《高光谱遥感数据的DNA计算分类》 865-878 1-4 第14卷, 第5期 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102445712A (en) * | 2011-11-22 | 2012-05-09 | 成都理工大学 | Character window weighting related spectrum matching method facing rocks and minerals |
CN102663752A (en) * | 2012-04-11 | 2012-09-12 | 南京理工大学 | SAM weighted KEST hyperspectral anomaly detection algorithm |
CN102663752B (en) * | 2012-04-11 | 2014-10-15 | 南京理工大学 | SAM weighted KEST hyperspectral anomaly detection algorithm |
CN104036293A (en) * | 2014-06-13 | 2014-09-10 | 武汉大学 | Rapid binary encoding based high resolution remote sensing image scene classification method |
CN104036293B (en) * | 2014-06-13 | 2017-02-22 | 武汉大学 | Rapid binary encoding based high resolution remote sensing image scene classification method |
CN104331482A (en) * | 2014-11-07 | 2015-02-04 | 武汉大学 | Method and system for hyper-spectrum matching based on feature index |
CN104331482B (en) * | 2014-11-07 | 2018-01-12 | 武汉大学 | The ultraphotic spectrum matching process and system of feature based index |
CN104867154A (en) * | 2015-05-29 | 2015-08-26 | 华中科技大学 | Remote-sensing image quality evaluation method based on gradient |
CN107798286A (en) * | 2017-07-13 | 2018-03-13 | 西安电子科技大学 | High spectrum image evolution sorting technique based on marker samples position |
CN107798286B (en) * | 2017-07-13 | 2021-01-26 | 西安电子科技大学 | Hyperspectral image evolution classification method based on labeled sample position |
CN111062888A (en) * | 2019-12-16 | 2020-04-24 | 武汉大学 | Hyperspectral image denoising method based on multi-target low-rank sparsity and spatial-spectral total variation |
CN111062888B (en) * | 2019-12-16 | 2022-02-15 | 武汉大学 | Hyperspectral image denoising method based on multi-target low-rank sparsity and spatial-spectral total variation |
CN112598265A (en) * | 2020-12-18 | 2021-04-02 | 武汉大学 | Decoupling risk estimation-based rapid detection method for hyperspectral pine nematode disease of unmanned aerial vehicle |
CN112598265B (en) * | 2020-12-18 | 2022-06-07 | 武汉大学 | Decoupling risk estimation-based rapid detection method for hyperspectral pine nematode disease of unmanned aerial vehicle |
CN112434685A (en) * | 2021-01-28 | 2021-03-02 | 中国科学院空天信息创新研究院 | Remote sensing image information fine extraction method based on earth surface elements |
CN112434685B (en) * | 2021-01-28 | 2021-05-28 | 中国科学院空天信息创新研究院 | Remote sensing image information fine extraction method based on earth surface elements |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102073882A (en) | Method for matching and classifying spectrums of hyperspectral remote sensing image by DNA computing | |
CN111199343B (en) | Multi-model fusion tobacco market supervision abnormal data mining method | |
CN111161793B (en) | Stacking integration based N in RNA 6 Method for predicting methyladenosine modification site | |
Zeebaree et al. | Combination of K-means clustering with Genetic Algorithm: A review | |
Espejo et al. | A survey on the application of genetic programming to classification | |
CN106815492B (en) | A kind of automatic method of bacterial community composition and diversity analysis for 16S rRNA gene | |
CN105929690B (en) | A kind of Flexible Workshop Robust Scheduling method based on decomposition multi-objective Evolutionary Algorithm | |
CN110990718B (en) | Social network model building module of company image lifting system | |
CN111832949B (en) | Construction method of equipment combat test identification index system | |
CN109165672A (en) | A kind of Ensemble classifier method based on incremental learning | |
CN113344019A (en) | K-means algorithm for improving decision value selection initial clustering center | |
CN104809476A (en) | Multi-target evolutionary fuzzy rule classification method based on decomposition | |
CN104966106A (en) | Biological age step-by-step predication method based on support vector machine | |
CN103164631B (en) | A kind of intelligent coordinate expression gene analyser | |
CN116601715A (en) | Drug optimization through active learning | |
CN115809601A (en) | Sedimentary rock structure background distinguishing method | |
Chakarverti et al. | Prediction analysis techniques of data mining: a review | |
CN109801681B (en) | SNP (Single nucleotide polymorphism) selection method based on improved fuzzy clustering algorithm | |
CN112528554A (en) | Data fusion method and system suitable for multi-launch multi-source rocket test data | |
CN114093420B (en) | XGboost-based DNA recombination site prediction method | |
CN114757433A (en) | Method for quickly identifying relative risk of drinking water source antibiotic resistance | |
CN112926640A (en) | Cancer gene classification method and equipment based on two-stage depth feature selection and storage medium | |
Gatnar | Tree-based models in statistics: Three decades of research | |
Bhat et al. | OTU clustering: A window to analyse uncultured microbial world | |
CN115995262B (en) | Method for analyzing corn genetic mechanism based on random forest and LASSO regression |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20110525 |