CN102298705A - Method for analyzing influence of landscape characteristic on remote sensing classification patch accuracy - Google Patents
Method for analyzing influence of landscape characteristic on remote sensing classification patch accuracy Download PDFInfo
- Publication number
- CN102298705A CN102298705A CN2011102303109A CN201110230310A CN102298705A CN 102298705 A CN102298705 A CN 102298705A CN 2011102303109 A CN2011102303109 A CN 2011102303109A CN 201110230310 A CN201110230310 A CN 201110230310A CN 102298705 A CN102298705 A CN 102298705A
- Authority
- CN
- China
- Prior art keywords
- spot
- classification
- data
- remote sensing
- precision
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a method for analyzing the influence of a landscape characteristic on remote sensing classification patch accuracy. The method comprises the following steps of: 1, acquiring data, including performing data standardization processing on an original image; 2, performing image recognition on the data obtained in the step 1, including classifying and classification post-processing, wherein the classifying process comprises determining a classification patch; 3, computing the relativity between the classification patch and a true value patch on the data obtained by recognizing in the step 2 and performing regression curve fitting to establish a regression model, including defining a landscape index representing the landscape characteristic; and 4, counting and verifying the regression model established in the step 3, and evaluating the classification error of landscape index expression simultaneously. In the method, the space characteristic of remote classification is taken as the theoretical basis of accuracy evaluation and analysis, and the classification accuracy is described and expressed by adopting the landscape characteristic of the remote sensing classification patch, so that basis and guidance are provided for relevant application and research of a land cover thematic map.
Description
[technical field]
The present invention relates to the remote sensing image classification field, particularly a kind of view feature of analyzing is to influence of remote sensing classification chart spot precision and the method for estimating.
[background technology]
Remote sensing classification thematic map is widely used in many fields, cover/utilize variation monitoring, the addressing of animal habitat ground, hydrological analysis, venture analysis and survey of natural resources as the soil, often be used to describe aspect (Stehman andCzaplewski, 1998 such as space distribution that the soil covers and form, estimation soil area coverage; Smith et al., 2002).The precision of thematic map and error profile directly have influence on range of application and effect (Foody, 2002 of thematic map; Smith et al., 2002; Smith et al., 2003).Therefore, the source of overall understanding soil covering thematic map error, space distribution scope cover the soil based on remote sensing classification thematic map and use have great importance (Congalton and Green, 1993; Congalton and Green, 1999; Yang et al., 2000; Shao et al., 2001).Soil covering precision expression accurately is that the soil covers the important component part that thematic map is used with estimating.From 1980, the evaluation of remote sensing classification quality was experienced from qualitative to quantitative, research process from non-position difference to differences in spatial location.Confusion matrix is widely used in the result of comprehensive description precision evaluation, each class is layouted at random calculate the mean accuracy of non-position difference.Such as, in the classification thematic map under the ratio of certain class atural object and the truth difference between the ratio of this atural object as the overall accuracy (Foody, 2002) of this atural object.Whole thematic map is counted as a space cell, and the precision of the nonspecific position that confusion matrix obtains can be counted as precision (Stehman etal., 2007 that the soil covers whole unit in the composition; Stehman, 2009a, b).Confusion matrix can obtain the quantitative target of remote sensing nicety of grading, divides sum of errors wrong error, producer's precision and user's precision (Congalton and Green, 1999 divided as overall accuracy, Kappa coefficient, leakage; Smith et al., 2002).But along with going deep into of research, confusion matrix is proved the quality of expression remote sensing classification that can not be complete.Foody thinks that in fact Kappa coefficient calculations process has over-evaluated chance coincidence, causes the overall nicety of grading of remote sensing to be underestimated.People such as Stehman research thinks, what the overall precision of confusion matrix was beyond expression the uncertainty of remote sensing classification and error has inclined to one side distribution (Koukoulas and Blackburn, 2001; Foody, 2002).
The generation factor of remote sensing error in classification has a lot, comprises mainly that atural object characteristic, sensor resolution, spectral space are obscured, preprocessing process and sorting technique etc. (Campbell, 1983).Because view features such as soil cover type heterogeneity is higher, the atural object patch is less, the atural object form is intricate all can cause the error of remote sensing classification, therefore, atural object view feature can reflect the inclined to one side distribution of having of remote sensing error in classification (Moisen et al., 1994).In addition, because the spectral characteristic that pixel reflected is subject to the influence of peripheral pixel spectrum, cause different object spectrum stacks and hybrid region pixel error in classification to increase (Townshend et al., 2000), thereby there is the wrong pixel (Campbell, 1987) that divides in the zone of transition between the different atural objects.Congalton (1988) utilizes forest land-non-forest land thematic map to carry out the inclined to one side distributional analysis of having of error, has verified the existence of this rule.Many subsequently scholars have estimated the relation (Wickham et al., 1997) of pixel present position and nicety of grading qualitatively based on experience and experiment.Relation between Smith and Stehman quantitative test landscape map spot size, pixel classification heterogeneity and the pixel error in classification, there are certain relation really in proof error in classification and view variable (figure spot size, heterogeneity), and then utilize the space characteristics factor to express possibility distribution (Smith et al., 2002 of error in classification; Smith et al., 2).But, these research all with single pixel as research object, only consider the influence that the pixel position may be caused, ignored the shape facility and the space distribution architectural characteristic of atural object itself.Common experience thinks that in the remote sensing classification results, classification chart shape of spot, scale distribute and may cause error profile to have bigger difference.The pool is high by utilizing the layering sign of classification chart spot degree of fragmentation as the remote sensing sampling recklessly, reflects the degree of fragmentation and the relation between the nicety of grading (Hu Tangao, 2010) of the spot of publishing picture to a certain extent.
[summary of the invention]
Cover the deficiency of thematic map precision evaluation method at traditional soil, the present invention is cell formation degree of fragmentation, area index with remote sensing classification chart spot, select two test blocks of winter wheat scale, broken plantation to conduct a research, inquire into the relation between view index and the nicety of grading, further deeply inquire into the theoretical foundation that the space characteristics that adopts the remote sensing classification carries out precision evaluation and analysis, proposed a kind of analytical approach that remote sensing classification chart spot precision is influenced based on the view feature.The present invention is intended to adopt the view feature of remote sensing classification chart spot that nicety of grading is described and expresses, thereby for covering the relevant application of thematic map with the soil and research provides foundation and guidance.
The view feature that the present invention proposes may further comprise the steps the analytical approach of remote sensing classification chart spot precision influence:
Step 1, obtain data, comprise that raw video is carried out data normalization to be handled, simultaneously the high score data are carried out visual interpretation, obtain true value soil cover data;
Step 2, the data that step 1 is obtained are carried out remote sensing recognition, comprise classification and classification aftertreatment, wherein include determining of classification chart spot in assorting process;
Thereby step 3, step 2 is carried out classification results data and true value data that pre-service obtains carry out relatedness computation, set up regression model and finish the regression curve match, comprising landscape index definition at the view feature;
Step 4, the model that step 3 is set up are tested, and the error in classification that the view index table is reached is estimated simultaneously.
Preferably, the data in the above-mentioned steps one comprise two kinds, and a kind of is 20m resolution spot data, and another kind is 2.4m resolution Quick Bird data, handle for first kind of The data data normalization, carry out visual interpretation for second kind of data.
Preferably, the pre-service of above-mentioned steps two comprises carries out the SVM supervised classification to the data after the standardization, obtains classification results, and then classify aftertreatment and binaryzation obtain the thematic map of classifying, and then carries out figure spot vector quantization again, obtains vector quantization figure spot.
Preferably, above-mentioned 2.4m resolution Quick Bird data are carried out obtaining with reference to true value behind the visual interpretation, for the curve fitting process of subsequent step three provides reference data.
Above-mentioned vector quantization figure spot is carried out landscape index calculate, wherein landscape index comprises figure spot area, figure spot degree of fragmentation, and then the relation between analytical calculation precision and the landscape index.
The reference true value of above-mentioned visual interpretation gained is carried out the calculating of figure spot precision, and then analyze the relation between nicety of grading and the landscape index.
Precision and landscape index that aforementioned calculation is obtained carry out correlation analysis, carry out curve fitting by setting up regression model, fitting result is carried out significance test, as then setting up regression model by check, do not pass through as check, then carry out match by regression model again, up to can be by significance test.
Preferably, above-mentioned steps four is specially carries out regression model to the result by significance test and sets up, and definite best curve regression model, further model is tested, thereby obtains the evaluation result of landscape index.
The analytical approach of utilizing the view feature to the influence of remote sensing classification chart spot precision provided by the invention, cover the deficiency of thematic map precision evaluation method at traditional soil, with remote sensing classification results figure spot is the cell formation degree of fragmentation, area index, select the plantation of winter wheat scale, two survey regions of broken plantation, inquire into the relation between degree of fragmentation and the nicety of grading, further deeply inquired into the theoretical foundation that the space characteristics that adopts remote sensing classification chart spot carries out precision evaluation and analysis, adopt the view feature of remote sensing classification chart spot that nicety of grading is described and expresses, thereby for covering the relevant application of thematic map with the soil and research provides foundation and guidance.
[description of drawings]
Fig. 1 is the classification chart spot of different scale;
Fig. 2 is the study area synoptic chart;
Fig. 3 is analytical approach experimental design of the present invention and process flow diagram;
Fig. 4 is figure spot nicety of grading evaluation method figure;
Fig. 5 a is that Daxing District figure spot precision and figure spot degree of fragmentation, figure spot area concern scatter diagram;
Fig. 5 b is that Tongzhou District figure spot precision and figure spot degree of fragmentation, figure spot area concern scatter diagram;
Fig. 6 a is Daxing District nicety of grading and figure spot area model regression fit figure;
Fig. 6 b is Daxing District nicety of grading and degree of fragmentation model regression fit figure;
Fig. 6 c is Tongzhou District nicety of grading and figure spot area model regression fit figure;
Fig. 6 d is Tongzhou District nicety of grading and degree of fragmentation model regression fit figure;
Fig. 7 a is Daxing District nicety of grading and figure spot area model regression fit figure;
Fig. 7 b is Daxing District nicety of grading and degree of fragmentation model regression fit figure;
Fig. 7 c is Tongzhou District nicety of grading and figure spot area model regression fit figure;
Fig. 7 d is Tongzhou District nicety of grading and degree of fragmentation model regression fit figure.
[embodiment]
The present invention is further described below in conjunction with description of drawings and embodiment.
At first, to error profile in the remote sensing classification and landscape index definition carrying out simple declaration.The factor that remote sensing nicety of grading and remote sensing error in classification produce has a lot, error, classification type mistake error and systematic error (operate miss) etc. that the error pattern that mainly comprises has mixed pixel to cause, and model formation is expressed as follows:
ξ
accuracy＝F
unction(E
mixed-pixel，E
system，E
misclassification，E
others) (1)
Wherein: ξ
AccuracyExpression remote sensing nicety of grading; E
Mixed-pixelThe error that the expression mixed pixel causes; E
SystemThe expression systematic error; E
MisclassificationThe presentation class type of error; E
OthersRepresent other error.
The error that the present invention mainly studies mixed pixel and caused.Because the mixed pixel that factors such as the restriction of sensor resolution and object spectrum are overlapping produce mainly is present in the intersection of different classes of atural object, it is to be the unit with the pixel to evaluations of layouting at random of classification thematic map that the tradition confusion matrix carries out precision evaluation, calculates the mean accuracy of the thematic map of classifying.Random sampling evaluation method based on point has the characteristic of non-position (stochastic distribution), but in fact remote sensing thematic map error in classification is not a stochastic distribution.People such as Smithand Stephen have inquired into error profile (Smith et al., 2002 of pixel position difference; Smithet al., 2003; Herrmann et al., 2005; Stehman and Wickham, 2006; Stehman 2009a), has confirmed the spatial distribution differences of remote sensing error in classification.Tradition is based on the precision evaluation of pixel, and its variable is a binary response variable, promptly to (1) and wrong (0).The space characteristics that the present invention is based on image classification figure spot carries out precision evaluation, not only comprise the position to mistake, comprise the correct ratio of figure spot simultaneously.Fig. 1 has discussed the figure spot situation from 1 pixel (per-pixel) to 6 pixels (multi-pixel), and the classification chart spot is formed with eight neighborhoods rule, and wherein the numeral below each figure spot is indicated the number of pixel common edge in this figure spot pixel set.Under the prerequisite of same pixel number figure spot, pixel common edge number reduces step by step, and along with the minimizing of common edge, independent pixel on the figure spot (do not have the pixel of common edge with periphery, be 8 neighborhoods with other pixels) increases gradually.According to the characteristic of remote sensing classification, pixel gathers more, is the rule schema spot, and then the nicety of grading of remote sensing is high more.Along with increasing gradually of independent pixel on the figure spot, then be irregular on the figure spot, thereby cause the quantity of mixed pixel on the figure spot to increase, cause the precision of classification results to reduce.
The present invention only considers that mixed pixel is to the influence of classification chart spot precision in the thematic map classification, and nicety of grading is correlated with independent variable as shown in Equation 2:
y＝f(x
1，x
2，…，x
i，…x
n) (2)
Wherein: x
iExpression landscape index variable is as indexs such as figure spot area, figure spot degree of fragmentation, special heterogeneity, figure spot girths; Y presentation class figure spot precision.
Whether meet linearity, nonlinear relationship about predicting between landscape index and the error in classification, therefore, need carry out linear regression and curve respectively to two variablees and estimate, reach the optimum regression model of determining the independent variable factor and dependent variable classification chart spot precision.
Degree of fragmentation is used to the research of ecological landscape index the earliest, is used for determining the shape and square or circular degree of closeness or integrated degree (Jaeger, 2000 of group; Zurita-Milla et al., 2011).The present invention uses for reference this thought, and express the shape facility of classification chart spot: mixed pixel mainly concentrates on the i.e. fringe region of figure spot of different atural object intersections, is significant positive correlation with classification chart spot girth; Simultaneously, figure spot size is significant positive correlation (Smith et al., 2002 with nicety of grading; Smith et al., 2003; Herrmann et al., 2005).Therefore, comprehensively the feature of the two is introduced view degree of fragmentation index (according to the characteristics of remote sensing classification, isolated pixel is subjected to the influence of peripheral pixel spectral information, error in classification maximum, then isolated pixel degree of fragmentation maximum, otherwise degree of fragmentation reduction), is defined as follows:
Wherein, F is the degree of fragmentation index, and P is a figure spot girth, and A is a figure spot area.The degree of fragmentation index has reflected the shape facility of figure spot to a certain extent, and the degree of fragmentation scope is between [0,1], the big more figure shape of spot that then shows is irregular more, more little then presentation graphs shape of spot is regular more, for figure spot with the number pixel, and foursquare degree of fragmentation minimum.Table 1 has been expressed by 1 to 6 pixel figure spot under the prerequisite of different number common edge, the result of calculation of classification chart spot degree of fragmentation.From the figure shape of spot as can be seen, figure spot degree of fragmentation is big more, and the isolated pixel on the figure spot is many more, reflects that to a certain extent potential mixed pixel increases.
Classification chart spot degree of fragmentation under the different pixel yardsticks of table 1
Table1?fragmentation?degree?with?multi-scale?pixels?collection
As shown in Figure 2, be the experimental study flow process that the invention process is carried out, this experimental study is selected Daxing District, Beijing and Tongzhou District, sees Fig. 2.Study area one is positioned at the area, northeast of Tongzhou District, and the winter wheat planting area concentrates on this west and south, zone relatively, presents the plantation feature of extensiveization, is suitable for the Classification and Identification of remote sensing.Daxing District winter wheat pattern of farming complexity, plot fragmentation, this accurately discerns for winter wheat remote sensing and causes very big difficulty (Hu Tangao, 2010).Therefore, the present invention chooses between Tongzhou District and Daxing District these two plantations tangible regional analysis landscape index of feature difference and the classification chart spot precision and concerns to have very strong representativeness.
Phenology information according to Beijing area winter wheat, the spot data (pixel resolution 20m) of selecting on April 27th, 2006 for use is used for the extraction of study area winter wheat thematic map information as raw video, select for use on May 2nd, 2006 Tongzhou District and on April 22nd, 2006 Daxing District QuickBird data (pixel resolution is 2.4m) visual interpretation obtain winter wheat area as the reference true value.At first, the SPOT raw video is carried out standardization (geometric correction and projection conversion etc.); Secondly, the data after handling are carried out the SVM supervised classification, obtain the classification results of winter wheat; Then, the classification results raster data is carried out eight neighborhood search realize that grid-arrow transforms, generate winter wheat classification chart spot image.
As shown in Figure 3, for the view feature to the influence of remote sensing classification chart spot precision and the idiographic flow of expression analysis, may further comprise the steps:
Step 1, obtain data, comprise and raw video carried out data normalization is handled and two kinds of methods of visual interpretation processing;
Step 2, the data that step 1 is obtained are carried out pre-service, wherein data comprise two kinds, a kind of is 20m resolution spot data, and another kind is 2.4m resolution QuickBird data, handles for first kind of The data data normalization, pretreated detailed process comprises carries out the SVM supervised classification to the data after the standardization, obtain classification results, then classify aftertreatment and binaryzation obtain the thematic map of classifying, then carry out figure spot vector quantization again, obtain vector quantization figure spot.2.4m resolution QuickBird data are carried out obtaining with reference to true value behind the visual interpretation, for the curve fitting process of subsequent step three provides the true variable data.
Thereby step 3, step 2 is carried out the data that remote sensing recognition obtains carry out relatedness computation, set up regression model and finish the regression curve match, comprising landscape index definition at the view feature; Specifically vector quantization figure spot is carried out landscape index and calculate, wherein landscape index comprises figure spot area, figure spot degree of fragmentation, and then the relation between analytical calculation precision and the landscape index.Specifically precision and the landscape index that calculates carried out correlation analysis, carry out curve fitting by setting up regression model, fitting result is carried out significance test, as then setting up regression model by check, do not pass through as check, then reselect other regression models and carry out match, up to can be, thereby determine optimum regression model by significance test.
Step 4, the model that step 3 is set up are tested, and simultaneously landscape index are estimated.Specifically the result by significance test is carried out regression model and set up, and set up the best curve regression model, further model is tested, thereby obtain the error assessment result.
Tradition is a correctness of judging single pixel classification based on the evaluation of pixel, and classification correctly is designated as 1, and classification error is designated as 0, for the pixel that is in different atural object intersections, is difficult to judge its realistic accuracy.For the figure spot, traditional judgement classification results correctness is the situation of realistic accuracy that is beyond expression, so the nicety of grading that adopts the correct regional shared ratio of classification to come the evaluation map spot.As shown in Figure 4, black surround part presentation graphs spot true value, red frame table shows the classification chart spot, hatched example areas is represented the two overlapping region.In the remote sensing classification, what precision information was contained is not only position attribution, comprises overlapping space segment simultaneously.Among Fig. 4, expressed the relation between classification chart spot and the true value figure spot: (a) expression true value figure spot; (b) the presentation class result overlaps completely with true value figure spot; (c)-(e) presentation class figure spot and figure spot true value part is overlapping.Be the influence to nicety of grading of the degree of fragmentation of evaluation map spot, the present invention as evaluation unit, is a scope to divide the buffer zone of pixel in one on the true value figure spot with true value figure spot, divides the figure spot precision of image in the evaluation.
General, the evaluation method of classification chart spot precision embodies with following formula:
Wherein, k represents the classification of atural object; λ
KkExpression overlapping region area; λ
K+The expression figure spot area is with reference to true value; λ
+ kThe presentation class figure spot area; UA represents user's precision; PA represents producer's precision; The ratio of RA presentation class result and true value area considers that true variable can not be 0, therefore is in the position of denominator, and span is [0,8], approaches 1 degree expression nicety of grading, and more near 1, then the precision of presentation class figure spot is high more; TA presentation class figure spot precision for guaranteeing that span is [0,1], is then handled RA, and value then is defined as 2 above 2 part, and the presentation class result overlaps with true value figure spot when the TA value is 1, and precision reaches 100%.
Next be that regression model is carried out match, overall accuracy TA according to classification chart spot precision evaluation calculating chart spot, classification results figure spot is calculated landscape index, obtain figure spot area, figure spot degree of fragmentation, 3 groups of results of figure spot precision respectively, by three-dimensional scatter plot distributions, its characteristic distributions as shown in Figure 5:
Fig. 5 a and 5b have represented the distribution scatter diagram of Daxing District, Tongzhou District figure spot nicety of grading and view variable respectively.Colored point is found out from the space: red yellow dots will precision is lower, and majority is distributed in degree of fragmentation and reaches the less zone of figure spot area more greatly; Figure spot area is big, the less part of degree of fragmentation, and nicety of grading presents high-precision blueness more.
As can be seen from Figure 5, there is certain rules in the distribution spatially of remote sensing error in classification, but has certain fluctuation for single figure spot, and is individual representative relatively poor.Therefore, figure spot index is taked the mode of classified statistics, promptly carry out segmentation and constitute one group, show to have the global feature and the precision of this group representative preferably by statistical indicators such as average, standard deviation and degree of confidence in the calculating group according to specific view index.At last, select suitable regression model, to carrying out regression fit between figure spot nicety of grading and the view characteristic index.
The regression relation model is carried out match, and the normal at present regression model that adopts comprises: monobasic linear model, quadratic function model, function of functions model, growth function model, logarithmic function model, sigmoid curve, exponential Function Model, power function model and logistic function model etc.According to Fig. 5 characteristics that the space of points distributes of loosing, each view variable is carried out the classified statistics of a fixed step size, thereby determine the relational model of this view variable and figure spot precision, select regression analysis model, and the curve of regression fit is carried out the parameter estimation of model, compare R
2, F test value, probable value etc. together.
Fig. 6 has represented nicety of grading and diffusing some statistical relationship that divides into groups of different landscape index in the test block, and the standard deviation (SD) in this landscape index group is defined as error line, and a, b and c, d have expressed the relation between Daxing District, Tongzhou District landscape index and the precision respectively.As can be seen from Figure 6, there are certain correlationship in nicety of grading and landscape index: nicety of grading and figure spot area present obvious positive correlation, and nicety of grading and degree of fragmentation present tangible negative correlation.Because having been carried out classified statistics, the point that looses is used for reflecting its group characteristics, so the standard deviation in the group can reflect the precision possible range of this index correspondence, there is following relation: the figure spot that area is less, mean accuracy is lower, group internal standard difference is bigger, and the nicety of grading otherness is also bigger.Along with the increase of figure spot area, its nicety of grading improves thereupon, and nicety of grading is comparatively stable; The figure spot that degree of fragmentation is bigger, its nicety of grading fluctuation is bigger, and along with reducing of degree of fragmentation, its nicety of grading improves thereupon.
Correlationship between nicety of grading and the figure spot landscape index is selected best fitting function model, the best fit model of figure spot area and precision mutual relationship is a S type curve model, figure spot degree of fragmentation index and figure spot precision optimum fit curve meet the cubic function model, and functional form is as follows:
Monobasic linear model: y=b
0+ b
1x
S type curve model:
Cubic function model: y=b
0+ b
1X+b
2x
2+ b
3x
3
The result that curve fitting obtains selects the homing method match to area exponential sum degree of fragmentation index as shown in Figure 7, sets up the relation between Trendline expression view index and the nicety of grading.Table 3 has been represented model of fit and parameter estimation, significance test result respectively.
Table 2 figure spot area and nicety of grading model and parameter estimation
Table2?Model?Summary?and?Parameter?Estimates?of?Patch-area?and?Accuracy
* R Square represents R
2The value of statistic; F represents the F test value; Df1, df2 represents degree of freedom; Sig. represent level of significance, probable value promptly accompanies; Constant represents constant term; B1, b2, b3 represents regression parameter.
Table 3 degree of fragmentation and nicety of grading model and parameter estimation
Table3?Model?Summary?and?Parameter?Estimates?of?Fragmentation?and?Accuracy
* R Square represents R
2The value of statistic; F represents the F test value; Df1, df2 represents degree of freedom; Sig. represent level of significance, probable value promptly accompanies; Constant represents constant term; B1, b2, b3 represents regression parameter.
In two test blocks that alignment degree differs greatly, landscape index (figure spot area, degree of fragmentation) all is significantly relevant (Fig. 7: a-d with nicety of grading; Table 2,3): the best-fit of figure spot area and nicety of grading meets the sigmoid curve model, Daxing District R
2Reach 0.990, Tongzhou District R
2Reach 0.986, all present extremely relevant; The best-fit of degree of fragmentation and nicety of grading meets cubic function model (Cubic), Daxing District R
2Reach 0.965, Tongzhou District R
2Reach 0.970, also all present extremely correlativity.
Cubic fitting model has the significant goodness of fit, and has all passed through significance test.F value in table 2, the table 3, the goodness of fit of reflection regression equation.The F statistical value is remarkable more, the goodness of fit of regression equation is also just high more, from table 2, can obtain: in the regression equation of figure spot area and nicety of grading, two study area linear fit F values are 19.253 and 25.136, the F of sigmoid function match then reaches 1487.246 and 1724.208, the two all by the F check, have the significant goodness of fit, and the sigmoid function model fitting obviously is better than linear fit.As can be seen from Table 4: the two has also all passed through the F check, has reached the goodness of fit of regression equation.Wherein, the F value of linear fit and cubic curve model fitting degree are comparatively approaching.
All exist Daxing District (plantation is broken) higher R to be arranged for area index and degree of fragmentation index than Tongzhou District (scale plantation)
2And the F test value, i.e. the zone that plantation is broken, landscape index and nicety of grading matched curve have more excellent degree of fitting.
Inquired into the relation of view feature and figure spot nicety of grading among the embodiment provided by the invention.Experiment is studied Daxing District, Tongzhou District winter wheat classification chart spot at intermediate-resolution image classification result.Relation by quantitative research landscape index and nicety of grading can obtain, and on linear degree of correlation, degree of fragmentation index and nicety of grading significantly are negative correlation, the medium positive correlation of figure spot area and nicety of grading.Wherein, area index can adopt sigmoid curve model fitting expression figure spot nicety of grading trend, and related coefficient can reach more than 0.98; Degree of fragmentation index and nicety of grading present cubic function relation, and the match correlativity can reach more than 0.97, under 95% degree of confidence all by significance tests such as F, t.This paper is by the analysis to classification results figure spot planform feature, determine the distribution of classification results error, thereby quantitative expression the precision and the fiducial interval of each classification chart spot, as the classification thematic map follow precision index, be used to instruct the thematic map application relevant with precision.
Above content be in conjunction with concrete preferred implementation to further describing that the present invention did, can not assert that concrete enforcement of the present invention is confined to these explanations.For the general technical staff of the technical field of the invention, without departing from the inventive concept of the premise, can also make some simple deduction or replace, all should be considered as belonging to protection scope of the present invention.
Claims (8)
1. analyze the view feature to the method that remote sensing classification chart spot precision influences for one kind, it is characterized in that: may further comprise the steps:
Step 1, obtain data, comprise that raw video is carried out data normalization to be handled, simultaneously the high score data are carried out visual interpretation, obtain the true value reference data;
Step 2, the data that step 1 is obtained are carried out remote sensing recognition, comprise classification and classification aftertreatment, wherein include determining of classification chart spot in assorting process;
Step 3, step 2 is carried out classification results data and true value data that pre-service obtains carry out relatedness computation, set up regression model and finish the regression curve match, comprising the landscape index definition of view feature;
Step 4, the model that step 3 is set up are tested, and the error in classification that the view index table is reached is estimated simultaneously.
2. analysis view feature according to claim 1 is to the method for remote sensing classification chart spot precision influence, it is characterized in that: the data in the described step 1 comprise two kinds, a kind of is 20m resolution spot data, another kind is 2.4m resolution Quick Bird data, handle for first kind of The data data normalization, carry out visual interpretation for second kind of data, obtain soil cover data accurately.
3. analysis view feature according to claim 1 is to the method for remote sensing classification chart spot precision influence, it is characterized in that: the pre-service of described step 2 comprises carries out the SVM supervised classification to the data after the standardization, obtain classification results, then classify aftertreatment and binaryzation, obtain the thematic map of classifying, then carry out the vector quantization of figure spot again, obtain vector quantization figure spot.
4. analysis view feature according to claim 2 is to the method for remote sensing classification chart spot precision influence, it is characterized in that: described 2.4m resolution Quick Bird data are carried out obtaining with reference to true value behind the visual interpretation, for the curve fitting process of follow-up step 3 provides data.
5. analysis view feature according to claim 3 is to the method for remote sensing classification chart spot precision influence, it is characterized in that: vector quantization figure spot is carried out landscape index calculate, wherein landscape index comprises figure spot area, figure spot degree of fragmentation, special heterogeneity and figure spot girth, and then the relation between analysis meter nomogram spot precision and the landscape index.
6. analysis view feature according to claim 4 is characterized in that the method for remote sensing classification chart spot precision influence: the reference true value of visual interpretation gained is carried out the calculating of figure spot precision, and then analytical calculation precision and landscape index.
7. according to claim 5 or 6 described analysis view features method to the influence of remote sensing classification chart spot precision, it is characterized in that: precision and the landscape index that calculates carried out correlation analysis, set up the regression curve model, fitting result is carried out significance test, as then setting up regression model by check, do not pass through as check, then carry out match by other regression models again, up to determining suitable regression model by significance test.
8. analysis view feature according to claim 7 is to the method for remote sensing classification chart spot precision influence, it is characterized in that: described step 4 is specially carries out regression model foundation to the result by significance test, and set up the best curve regression model, further model is tested, thereby obtain trueness error evaluation result based on landscape index.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102303109A CN102298705A (en) | 2011-08-12 | 2011-08-12 | Method for analyzing influence of landscape characteristic on remote sensing classification patch accuracy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102303109A CN102298705A (en) | 2011-08-12 | 2011-08-12 | Method for analyzing influence of landscape characteristic on remote sensing classification patch accuracy |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102298705A true CN102298705A (en) | 2011-12-28 |
Family
ID=45359114
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011102303109A Pending CN102298705A (en) | 2011-08-12 | 2011-08-12 | Method for analyzing influence of landscape characteristic on remote sensing classification patch accuracy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102298705A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103077400A (en) * | 2012-12-26 | 2013-05-01 | 中国土地勘测规划院 | Land type information remote sensing automatic identification method supported by land use database |
| CN103674014A (en) * | 2013-12-10 | 2014-03-26 | 中国神华能源股份有限公司 | Positioning method and device for natural gas well |
| CN103971115A (en) * | 2014-05-09 | 2014-08-06 | 中国科学院遥感与数字地球研究所 | Automatic extraction method for newly-increased construction land image spots in high-resolution remote sensing images based on NDVI and PanTex index |
| CN105279223A (en) * | 2015-10-20 | 2016-01-27 | 西南林业大学 | Computer automatic interpretation method for remote sensing image |
| CN105608473A (en) * | 2015-12-31 | 2016-05-25 | 中国资源卫星应用中心 | High-precision land cover classification method based on high-resolution satellite image |
| CN106909899A (en) * | 2017-02-24 | 2017-06-30 | 中国农业大学 | A kind of analysis method and analysis system of wetland landscape evolution process |
| CN108346163A (en) * | 2018-02-09 | 2018-07-31 | 河南城建学院 | A kind of land use area reckoning method |
| CN108710864A (en) * | 2018-05-25 | 2018-10-26 | 北华航天工业学院 | Winter wheat Remotely sensed acquisition method based on various dimensions identification and image noise reduction processing |
| CN110175991A (en) * | 2019-05-24 | 2019-08-27 | 四川九洲北斗导航与位置服务有限公司 | View evaluation method and device based on high score remote sensing |
| CN110705449A (en) * | 2019-09-27 | 2020-01-17 | 佛山科学技术学院 | Land utilization change remote sensing monitoring analysis method |
| CN111915669A (en) * | 2020-08-03 | 2020-11-10 | 北京吉威空间信息股份有限公司 | Land survey linear ground object pattern spot method based on total amount control |
-
2011
- 2011-08-12 CN CN2011102303109A patent/CN102298705A/en active Pending
Non-Patent Citations (2)
| Title |
|---|
| 崔林丽: "《遥感影像解译特征的综合分析与评价》", 《中国优秀硕士学位论文全文数据库》 * |
| 黄颖: "《基于遥感与景观指数的土地利用/覆盖变化及格局分析》", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103077400B (en) * | 2012-12-26 | 2015-11-25 | 中国土地勘测规划院 | The ground category information remote sensing automatic identifying method that Land Use Database is supported |
| CN103077400A (en) * | 2012-12-26 | 2013-05-01 | 中国土地勘测规划院 | Land type information remote sensing automatic identification method supported by land use database |
| CN103674014A (en) * | 2013-12-10 | 2014-03-26 | 中国神华能源股份有限公司 | Positioning method and device for natural gas well |
| CN103674014B (en) * | 2013-12-10 | 2017-01-04 | 中国神华能源股份有限公司 | The localization method of the natural gas well and device |
| CN103971115A (en) * | 2014-05-09 | 2014-08-06 | 中国科学院遥感与数字地球研究所 | Automatic extraction method for newly-increased construction land image spots in high-resolution remote sensing images based on NDVI and PanTex index |
| CN103971115B (en) * | 2014-05-09 | 2017-05-10 | 中国科学院遥感与数字地球研究所 | Automatic extraction method for newly-increased construction land image spots based on NDVI and PanTex index |
| CN105279223A (en) * | 2015-10-20 | 2016-01-27 | 西南林业大学 | Computer automatic interpretation method for remote sensing image |
| CN105608473B (en) * | 2015-12-31 | 2019-01-15 | 中国资源卫星应用中心 | A kind of high-precision land cover classification method based on high-resolution satellite image |
| CN105608473A (en) * | 2015-12-31 | 2016-05-25 | 中国资源卫星应用中心 | High-precision land cover classification method based on high-resolution satellite image |
| CN106909899A (en) * | 2017-02-24 | 2017-06-30 | 中国农业大学 | A kind of analysis method and analysis system of wetland landscape evolution process |
| CN106909899B (en) * | 2017-02-24 | 2019-11-05 | 中国农业大学 | A kind of analysis method and analysis system of wetland landscape evolution process |
| CN108346163A (en) * | 2018-02-09 | 2018-07-31 | 河南城建学院 | A kind of land use area reckoning method |
| CN108710864A (en) * | 2018-05-25 | 2018-10-26 | 北华航天工业学院 | Winter wheat Remotely sensed acquisition method based on various dimensions identification and image noise reduction processing |
| CN108710864B (en) * | 2018-05-25 | 2022-05-24 | 北华航天工业学院 | Winter wheat remote sensing extraction method based on multi-dimensional identification and image noise reduction processing |
| CN110175991A (en) * | 2019-05-24 | 2019-08-27 | 四川九洲北斗导航与位置服务有限公司 | View evaluation method and device based on high score remote sensing |
| CN110705449A (en) * | 2019-09-27 | 2020-01-17 | 佛山科学技术学院 | Land utilization change remote sensing monitoring analysis method |
| CN111915669A (en) * | 2020-08-03 | 2020-11-10 | 北京吉威空间信息股份有限公司 | Land survey linear ground object pattern spot method based on total amount control |
| CN111915669B (en) * | 2020-08-03 | 2024-04-05 | 北京吉威空间信息股份有限公司 | Total quantity control-based land survey linear ground object pattern spotting method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102298705A (en) | Method for analyzing influence of landscape characteristic on remote sensing classification patch accuracy | |
| Seress et al. | Quantifying the urban gradient: a practical method for broad measurements | |
| Panagos et al. | Estimating the soil organic carbon content for European NUTS2 regions based on LUCAS data collection | |
| Münzinger et al. | Mapping the urban forest in detail: From LiDAR point clouds to 3D tree models | |
| Congalton | Thematic and positional accuracy assessment of digital remotely sensed data | |
| CN109766824B (en) | Active and passive remote sensing data fusion classification method based on fuzzy evidence theory | |
| CN102354328A (en) | Leaf area index (LAI) product inversion method and system for global earth surface | |
| CN113360587B (en) | Land surveying and mapping equipment and method based on GIS technology | |
| Li et al. | An iterative-mode scan design of terrestrial laser scanning in forests for minimizing occlusion effects | |
| CN116030352B (en) | Long-time-sequence land utilization classification method integrating multi-scale segmentation and super-pixel segmentation | |
| Kathuria et al. | Development of an automated individual tree detection model using point cloud LiDAR data for accurate tree counts in a Pinus radiata plantation | |
| CN112668613A (en) | Satellite infrared imaging effect prediction method based on weather forecast and machine learning | |
| CN109034189A (en) | Forest type identification method based on high score remote sensing image | |
| Sumnall et al. | Analysis of a lidar voxel-derived vertical profile at the plot and individual tree scales for the estimation of forest canopy layer characteristics | |
| Xu et al. | Vegetation information extraction in karst area based on UAV remote sensing in visible light band | |
| CN113901348A (en) | Oncomelania snail distribution influence factor identification and prediction method based on mathematical model | |
| CN105894006A (en) | Space-time probability model rice remote sensing recognition method | |
| CN107657262B (en) | A kind of computer automatic sorting Accuracy Assessment | |
| Khairiah et al. | Monitoring model of payment for environmental service (PES) implementation in Cidanau Watershed with stands density approach | |
| Wei et al. | Automated mapping of landforms through the application of supervised classification to lidAR-derived DEMs and the identification of earthquake ruptures | |
| Förster et al. | Significance analysis of different types of ancillary geodata utilized in a multisource classification process for forest identification in Germany | |
| CN109060858A (en) | A kind of quantitative evaluation method of heavy metal-polluted soil concentration space forecast of distribution accuracy | |
| CN103955914A (en) | SAR image segmentation method based on random projection and Signature/EMD framework | |
| Apriyanto et al. | Examining the object-based and pixel-based image analyses for developing stand volume estimator model | |
| CN113344866A (en) | Point cloud comprehensive precision evaluation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20111228 |