CN105447452A - A remote sensing sub-pixel mapping method based on the spatial distribution characteristics of ground objects - Google Patents

Abstract

The invention discloses a remote sensing sub-pixel mapping method based on the spatial distribution characteristics of ground objects. Firstly, according to the spatial geometric characteristics of the ground objects, remote sensing images are divided into three ground object types: planar mode, linear mode and point mode. Secondly, under the assumption of spatial dependence, use the vector boundary-based sub-pixel mapping method to deal with surface features; use the linear feature template sub-pixel mapping method to deal with linear features; The meta-mapping method deals with point objects. Finally, the sub-pixel mapping results of the three spatial modes are mosaiced to obtain the sub-pixel mapping of the image. The present invention theoretically constructs a sub-pixel cartographic theoretical model capable of simultaneous point-line-surface feature and based on the spatial distribution characteristics of the feature, and has high simulation precision.

Description

A remote sensing sub-pixel mapping method based on the spatial distribution characteristics of ground objects

technical field

The invention relates to a sub-pixel mapping method, which belongs to the technical field of geospatial information.

Background technique

Remote sensing science and technology have been widely and deeply applied in fields such as land resource survey, hydrological process simulation, landscape ecology, agricultural yield estimation and disaster monitoring, forestry resource management, global climate change (Mei Anxin, 2001; Zhou Chenghu et al., 1999), high space Remote sensing imagery and its products with high, temporal and spectral resolutions are the development trend for further research and application in various fields (Zhao Yingshi, 2003). However, in the process of remote sensing image acquisition, due to the influence of the external environment and the limitations of the internal sensor itself, mixed pixels (MixedPixel) generally exist in the image, which limits the application of remote sensing images and restricts the extraction of remote sensing information (DeJongandvanderMeer, 2004; Fisher, 1997). Especially in the process of remote sensing classification, mixed pixels will cause errors and uncertainties in the results of traditional hard classification (Atkinson, 2009; Bai Yanchen and Wang Jinfeng, 2003; Ge Yong and Wang Jinfeng, 2003; Ge Yong and Li Sanping, 2008). In order to reduce the error caused by hard classification and specifying mixed pixels as a single category, the soft classification (SoftClassification) method was proposed and used a more scientific and reasonable way to express the classification results. Image) in the form of describing the area ratio of each object category in the pixel. Although the result of soft classification expresses more reasonable information than hard and soft classification, the spatial position of each object category in the mixed pixel is unknown, which brings new uncertainties (Tatemetal, 2002a; Ibrahimetal, 2005), that is, the uncertainty of the spatial location of sub-pixel category attributes. In order to take into account the advantages of soft classification methods that can scientifically and reasonably express the classification results and hard classification methods that can identify specific spatial locations, the sub-pixel mapping (or super-resolution mapping) , Sub-pixelMappingorSuper-resolutionMapping) concept was formally proposed and attracted the attention and attention of scholars (Atkinson, 1997).

Remote sensing sub-pixel mapping is to use the soft class results of low spatial resolution to locate the spatial distribution map of ground objects at a given arbitrary high spatial resolution within the mixed pixel. The main object of sub-pixel mapping research is mixed pixels, the reasons for which are complex and diverse, and the types of mixed pixels produced by different imaging conditions are also different. According to the different spatial distribution characteristics of the ground objects in the mixed pixel, it is generally accepted that a kind of standard is to divide it into two types (Woodcock and Strahler, 1987): (1) H-Type or H-Resolution refers to the surface feature spot The area of the block is larger than one pixel and the mixed pixel is generated at the junction of various ground objects. The ground objects in it are H-type ground objects with spatial correlation distribution characteristics, which can be approximately considered as surface objects ( AreaPattern, AP); (2) L-type mixed pixel (L-Type or L-Resolution) refers to the mixed pixel generated by the surface patch smaller than the size of a pixel area distributed in the pixel in a scattered and independent form, where The ground objects in are L-shaped objects with spatial heterogeneity distribution characteristics, which can be approximated as point pattern objects (PointPattern, PP). However, the three major geographical elements include not only the surface features in the H-type mixed pixel and the point-like features in the L-shaped mixed pixel, but also the spatial distribution characteristics of the surface and point features that are obviously different from those with fine details. Long-connected linear features (LinePattern, LP). Therefore, the sub-pixel mapping of linear ground objects is also one of the focuses that many scholars need to pay attention to and study.

After more than ten years of development and research, scholars at home and abroad have successively proposed many excellent sub-pixel mapping algorithms, most of which are mainly based on the maximum spatial correlation to simultaneously locate the mixed pixels in the entire image. The spatial distribution of point, line and surface objects in the L-shape mixed pixel is very little. Although the sub-pixel mapping algorithm based on the largest spatial correlation has a good processing effect on surface objects, it will cause the aggregation of point objects when dealing with point objects with heterogeneous distribution characteristics, and When dealing with linear features, it is often difficult to ensure the connectivity and spatial distribution characteristics of linear features, and it is easy to produce "sawtooth effect". Therefore, the development of sub-pixel mapping technology has prompted people to realize that only the sub-pixel mapping method based on the largest spatial correlation will restrict its further development and application, and it has also made people pay attention to and study the point and line inside the mixed pixel. Mapping problems of spatially distributed features. Therefore, while considering the sub-pixel mapping of surface objects in the H-type mixed pixel, it is necessary to attach great importance to and study the mapping of point-shaped and linear spatially distributed feature objects in the mixed pixel, and analyze and study them theoretically. The mechanism, properties and types of different types of mixed pixels produced by point, line and surface objects, and the processing technology that can simultaneously deal with the sub-pixel mapping of point, line and surface objects in mixed pixels, so as to make up for the current sub-image Insufficient meta-cartography research, further improving the accuracy of fine land cover classification maps at the sub-pixel scale.

Contents of the invention

The technical problem solved by the present invention is to overcome the deficiencies of the prior art, provide a remote sensing sub-pixel mapping method for the spatial distribution characteristics of ground objects, and theoretically construct a method based on the spatial distribution of ground objects that can simultaneously locate points, lines, and planes. The characteristic sub-pixel mapping theoretical model, the result of sub-pixel mapping has good visual effect and high precision.

First, for the fractional image C, the spatial distribution mode (area distribution, linear distribution and point distribution) of ground objects is judged by using the region growth model and shape index. Secondly, the surface object sub-pixel mapping method based on the boundary polygon is used to carry out sub-pixel mapping of the surface object of the fractional image; the linear template matching algorithm is used to perform sub-pixel mapping of the linear object of the fractional image; The spatial pattern consistency matching algorithm of ground objects is used for sub-pixel mapping of point-like ground objects in fractional images. Finally, the sub-pixel mapping results of surface features, point features and linear features are mosaiced in turn to obtain the sub-pixel mapping results of the entire fractional image C, and the null pixel is replaced by the feature type with the largest neighborhood frequency value.

The technical scheme of the present invention: a remote sensing sub-pixel mapping method based on the spatial distribution characteristics of ground objects, comprising the following steps:

Step 1. Preprocess the input remote sensing image, and then use the soft classification method to obtain the component proportion of each feature category in each pixel of the remote sensing image, which is called the abundance or fractional image; The category is three kinds of features, namely point features, linear features and area features;

Step 2. Use the region growth model to divide the fractional image into different regions, and use an ordered data storage structure to store the divided regions, so as to facilitate the calculation of the area, perimeter and boundary of each region;

Step 3, the type of object in the divided area: calculate the shape density index of the area of the fractional image after segmentation, and realize the division of point-shaped objects, linear objects and planar objects according to the corresponding discrimination threshold;

Step 4. According to the division results of the three types of ground objects in step 3, use the sub-pixel mapping method based on linear feature template matching for the mixed pixels of linear features, and use the sub-pixel mapping method based on vector boundaries for the mixed pixels of surface features. The pixel mapping method adopts the sub-pixel mapping method based on spatial correlation and model matching for the mixed pixels of point objects, and obtains the sub-pixel mapping results of point objects, linear objects, and surface objects;

Step 5: Mosaic the sub-pixel mapping results of point objects, linear objects, and planar objects, and finally obtain the sub-pixel mapping results.

The step 2 uses the region growing model to segment the fractional image. The basic idea of the region growing model is to combine pixels with similar properties, including gray level, texture, and gradient information to form a region, and finally achieve the purpose of image segmentation. The implementation process is as follows:

(1) Randomly select n pixels in the fractional image as seeds;

(2) Expand the area where the seed is located until all the pixels of the fractional image are divided into a specific area. For the neighboring pixels of the seed, if the difference between the pixel value and the pixel value of the seed is within the specified threshold ε, Then expand the neighborhood pixel to the area where the current seed is located.

In order to ensure the connectivity of linear features, the threshold ε is set to 1/S, and the sub-pixel mapping is obtained from the soft class results of low spatial resolution remote sensing images to obtain the remote sensing land classification map at a given high spatial resolution process, S is the magnification factor from low resolution to high resolution in sub-pixel mapping.

In the step 3, calculate the shape density index of the region of the fractional image after segmentation, and realize the division of point-like features, linear features and planar features according to the corresponding discrimination thresholds as follows;

I＝ρ ₁ ×S _shape +ρ ₂ ×D _density

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; \&times;</span>\sqrt{\mathrm{<span\; class="notranslate">\; A</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}$

Among them, I represents the shape density index; S _shape represents the shape index of the region, L represents the number of pixels on the border of the region, A represents the area of the region, D _density represents the density index of the region, and N represents the number of pixels in the outer rectangle of the region M is the number of rectangular boundary pixels outside the region, ρ ₁ and ρ ₂ represent the weight index, where ρ ₁ +ρ ₂ =1; if S≥2.3, D≤1.1, I≥1.6, the region is linear Otherwise, the area is divided into surface features; for the rest of the area, the pixels with a score greater than 1/S ² are divided into point features.

Described step 4, the method for sub-pixel mapping of linear features adopts the linear feature template matching method, including three steps of defining linear templates, matching linear templates and sub-pixel mapping of linear features, specifically as follows:

(1) Define linear templates: in order to succinctly represent as complete linear features as possible, the present invention utilizes 20 linear templates of 3 × 3 pixels, and the linear templates include straight line and curve types;

(2) Matching linear template: In order to find the most suitable linear object template in the mixed pixel, the present invention uses the correlation coefficient of the 3×3 pixel template and the 3×3 local window of the fractional image to represent their correlation, and calculates The formula is as follows.

${\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&times;</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}}$

Among them, F _jc represents the jth pixel of the C-th object score image, x, y represent the coordinates of the pixel, T _k represents the kth linear template, m, n represent the pixels in the 3×3 pixel linear template position, and stipulate that the coordinates of the upper left corner pixel are -1, -1, the present invention uses the linear template that makes r _jk the largest as the best template for the mixed pixel; (x+m, y+n) represents the jth Neighborhood pixels of the pixel, the fractional value of the fractional image of the C-type ground object;

(3) Sub-pixel mapping of linear features: at F _jc , the above two steps determine the best direction of linear features in the mixed pixel, that is, the direction of the best linear template, and define the upper left corner of pixel j The coordinates are (0, 0), and the coordinates of the lower right corner are (1, 1), so that the size of the linear template is equal to the mixed pixel j, then the coordinates of the pixels in the linear template are calculated by the following formula:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 6</span>}}\\ {\mathrm{<span\; class="notranslate">\; no</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 6</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}$

The pixels of each sub-pixel in the mixed pixel can be expressed by the following formula:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; S</span>}}\\ {\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; S</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; ...</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; S</span>}$

Among them, x, y are the number of rows and columns of sub-pixels, S is the magnification factor, that is, the multiple of image expansion, and the minimum Euclidean distance of 1 pixel from each sub-pixel to the best linear template _Tk is calculated, Arrange these minimum Euclidean distances in ascending order, and the surface object type of the first F _cj ×S ² sub-pixel is the C type surface object.

Said step 4 adopts the sub-pixel mapping method based on the vector boundary for the method of surface feature sub-pixel mapping and includes the following four steps:

(1) For each mixed pixel Use the vector boundary to extract the model to get the length and position of each line segment on the boundary of the mixed pixel;

(2) Connect each line segment generated in step (1) in a clockwise order to obtain the initial polygon boundary of the object C in the mixed pixel;

(3) Deleting the overlapping part of the initial polygon boundary obtained in step (2) and each line segment of step (1), that is, obtaining the final vector boundary;

(4) Use the ray discrimination method to assign values to the features within the vector boundary.

The step 4 adopts the sub-pixel mapping method based on spatial correlation and model matching for point-like features, and uses the Moran index to analyze the spatial distribution pattern of point-like features. The spatial pattern of the point-like feature changes with the size of the space range. The present invention uses a 3×3 pixel window to calculate the Moran index of the point-like feature. The specific process is as follows;

(1) For the fractional image of the C-type ground object type, firstly, the surface object sub-pixel mapping method is used to obtain the sub-pixel mapping image;

(2) For the point-like feature pixel j in the fractional image of the C-type feature type, calculate the Moran index value I of the local window where it is located; for the image obtained by the sub-pixel mapping method of the mixed pixel, calculate 3×3×S ² Moran’s exponent I’ of mixed pixel j;

(3) If |I'-I|≤θ, θ is the threshold for distinguishing whether the spatial distribution of pixel-level point-like objects and sub-pixel-like objects matches, then the sub-pixel mapping image is point-like The final result of the object; if |I'-I|≥θ, randomly adjust the attribute values of two pixels in the sub-pixel mapping image of the mixed pixel j, and recalculate the value of Moran's index until |I'-I| ≤ θ, the sub-pixel mapping results of the point-like features of the fractional images of the C-type feature types are obtained.

In step 5, the process of integrating and mosaicing the sub-pixel mapping results of point-like features, linear features, and area-like features to obtain the final sub-pixel mapping results is: firstly, the sub-pixel mapping of the area features The image is used as the background layer, and the sub-pixel mapping results of point objects and the sub-pixel mapping images of linear objects are superimposed in sequence; during the mosaic process, if the pixel value of the sub-pixel mapping image is null, use The 3×3 window traverses the neighborhood, and the category with the most occurrences is the category attribute of the pixel.

Compared with the prior art, the present invention has the advantages that: the present invention is a sub-pixel mapping method considering the spatial distribution mode of ground objects (area ground objects, linear ground objects and plane land objects). For the sub-pixel mapping of surface objects, it can ensure the maximization of the spatial correlation of surface objects, and can effectively improve the accuracy of sub-pixel mapping of surface objects; for the sub-pixel mapping of linear objects , which can ensure the connectivity of linear objects and produce relatively smooth boundaries of linear objects; for sub-pixel mapping of point objects, it can ensure the distribution pattern of point objects in the results of sub-pixel mapping Similarity to reference imagery improves cartographic accuracy. Compared with the traditional method, this algorithm starts from the perspective of geosciences, fully considers the spatial distribution mode of ground objects (area distribution, linear distribution and point distribution), and is based on the classification of ground objects in different spatial patterns, and has strong explanatory power. The result of sub-pixel mapping has good visual effect and high precision.

Description of drawings

Fig. 1 is main flowchart of the present invention;

Fig. 2 is a linear template diagram in the present invention, including straight lines and curve types, representing the direction of linear features in the real world.

Fig. 3 is the experimental data of the present invention, (a) is the standard false-color synthetic image of 400*400 picture element TM image, is a kind of common, in aspect remote sensing image processing such as vegetation, crops and land use, this is the most commonly used band combinations. (b) is a Google Earth image with higher spatial resolution and the same spatial extent as (a);

Figure 4 shows the fractional images of four kinds of ground features obtained by soft classification of TM images (in order of vegetation, waters, roads, and buildings). The pixel value of the fractional image is the proportion of the type C of the ground features in the mixed pixel. The highest value is 1, and the lowest value is 0;

Figure 5 shows the spatial distribution patterns of the regional objects after segmentation in the four types of surface object fractional images. The spatial distribution patterns of the objects include point distribution, linear distribution and planar distribution. Where (a) is vegetation, (b) is water, (c) is road, (d) is building;

Figure 6 is a comparison of different sub-pixel mapping algorithms, (a) is the SVM hard classification result of Google Earth image data, (b) is the sub-pixel mapping result of the SAM algorithm, (c) is the sub-pixel mapping of the HIIPD algorithm Results, (d) is the sub-pixel mapping result of the PSA algorithm, (e) is the sub-pixel mapping result of the LPSA algorithm, (f) is the sub-pixel mapping result of the MRF algorithm, (g) is the sub-pixel mapping result of the MAP algorithm The result of meta-mapping, (h) is the sub-pixel mapping result of the SPMv algorithm, (i) is the sub-pixel mapping method based on the distribution characteristics of spatial objects used in the present invention.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

As shown in Figure 1, the specific implementation process of the present invention is as follows:

Step 1. Perform geometric correction, atmospheric correction, and denoising preprocessing on the remote sensing image, and then softly classify the real remote sensing image according to the training samples obtained from the field survey to obtain the proportion of various ground object categories in each pixel , that is, the fractional image, which is used as the input data of the present invention. Since errors often exist in the soft classification process, in order to prevent the errors in the soft classification process from affecting sub-pixel mapping, the present invention uses high-resolution images to describe the specific implementation process of sub-pixel mapping. The present invention utilizes the high-resolution image of Google Maps as a reference image to verify the effectiveness of the sub-pixel mapping method.

Step 2. Use the region growing model to segment the fractional image. The specific process is as follows:

(1) Randomly select n pixels in the fractional image as seeds;

(2) Expand the area where the seed is located until all the pixels of the fractional image are divided into a specific area. For the neighboring pixels of the seed, if the difference between the pixel value and the pixel value of the seed is within the specified threshold ε, Then expand the neighborhood pixel to the area where the current seed is located. In order to ensure the connectivity of linear objects, the threshold ε is set to 1/S, and S is the amplification factor from low resolution to high resolution in sub-pixel mapping.

Step 3. Calculate the shape density index of the region after fractional image segmentation, and the calculation formula is as follows:

I＝ρ ₁ ×S _shape +ρ ₂ ×D _density

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; \&times;</span>\sqrt{\mathrm{<span\; class="notranslate">\; A</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}$

Among them, I represents the shape density index; S _shape represents the shape index of the region, L represents the number of pixels on the border of the region, A represents the area of the region, D _density represents the density index of the region, and N represents the number of pixels in the outer rectangle of the region M is the number of rectangular boundary pixels outside the region, ρ ₁ and ρ ₂ represent the weight index, where ρ ₁ +ρ ₂ =1; if S≥2.3, D≤1.1, I≥1.6, the region is linear Otherwise, the area is divided into surface features; for the rest of the area, the pixels with a score greater than 1/S ² are divided into point features.

Step 4. Use the corresponding sub-pixel mapping method for the three ground objects.

Firstly, for linear objects, the sub-pixel mapping method of linear object template matching is adopted. The specific process is as follows:

(1) Define linear templates: in order to succinctly represent as complete linear features as possible, the present invention utilizes 20 linear templates of 3 × 3 pixels, and the linear templates include straight line and curve types;

(2) Matching linear template: In order to find the most suitable linear object template in the mixed pixel, the present invention uses the correlation coefficient of the 3×3 pixel template and the 3×3 local window of the fractional image to represent their correlation, and calculates The formula is as follows.

${\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&times;</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}}$

Among them, F _jc represents the jth pixel of the C-th object score image, x, y represent the coordinates of the pixel, T _k represents the kth linear template, m, n represent the pixels in the 3×3 pixel linear template position, and stipulate that the coordinates of the upper left corner pixel are -1, -1, the present invention uses the linear template that makes r _jk the largest as the best template for the mixed pixel; (x+m, y+n) represents the jth Neighborhood pixels of the pixel, the fractional value of the fractional image of the C-type ground object;

(3) Sub-pixel mapping of linear features: For F _jc , the above two steps determine the best direction of linear features in the mixed pixel, that is, the direction of the best linear template, and define the upper left corner of pixel j The coordinates are (0, 0), and the coordinates of the lower right corner are (1, 1), so that the size of the linear template is equal to the mixed pixel j, then the coordinates of the pixels in the linear template are calculated by the following formula:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 6</span>}}\\ {\mathrm{<span\; class="notranslate">\; no</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 6</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}$

The pixels of each sub-pixel in the mixed pixel can be expressed by the following formula:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; S</span>}}\\ {\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; S</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; ...</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; S</span>}$

Among them, x, y are the number of rows and columns of sub-pixels, S is the magnification factor, that is, the multiple of image expansion, and the minimum Euclidean distance of 1 pixel from each sub-pixel to the best linear template _Tk is calculated, Arrange these minimum Euclidean distances in ascending order, and the surface object type of the first F _cj ×S ² sub-pixel is the C type surface object.

Secondly, the vector boundary-based sub-pixel mapping method for surface objects includes the following four processes:

(1) For each mixed pixel Use the vector boundary extraction model to obtain the length and position of each line segment on the boundary of the mixed pixel; (2) connect each line segment generated in step (1) clockwise to obtain the initial polygon of the C-type ground object in the mixed pixel Boundary; (3) delete the initial polygonal boundary that step (2) obtains and each line segment coincidence part of step (1), obtain the final vector boundary; assignment.

Finally, the sub-pixel mapping method of pattern matching is adopted for point objects. (1) Calculate the Moran index value I of the 3×3 local window where the fractional image mixed pixel j is located and the Moran index I' of the 3×3×S sub-pixels. (2) Compare the two Moran indices, if their difference is less than the specified threshold, the current sub-pixel image is the sub-pixel mapping result of point-like features, otherwise randomly adjust the sub-pixel mapping image of pixel j The attribute values of two pixels, recalculate the Moran index value, until the difference between the two Moran indices is less than the threshold, and the sub-pixel mapping image of point-like features is obtained.

Step 5: Integrate the sub-pixel mapping results of point objects, linear objects, and planar objects. The process of mosaicing to obtain the final sub-pixel mapping results is as follows: firstly, the sub-pixel mapping image of the surface features is used as the background layer, and the sub-pixel mapping results of the point features and the sub-pixel mapping results of the linear features are superimposed successively. Pixel cartographic image; in the process of mosaicing, if the pixel value of the sub-pixel cartographic image is null, a 3×3 window is used to traverse the neighborhood, and the category with the most occurrences is the category attribute of the pixel.

In order to compare the performance of the SPMs method proposed by the present invention and the traditional sub-pixel mapping method in the experiment, the experiment of 8 sub-pixel mapping methods was carried out in this process. The experimental results are as follows: in Figure 6, (b) is the sub-pixel mapping result of the SAM algorithm, (c) is the sub-pixel mapping result of the HIPD algorithm, (d) is the sub-pixel mapping result of the PSA algorithm, (e ) is the sub-pixel mapping result of LPSA algorithm, (f) is the sub-pixel mapping result of MRF algorithm, (g) is the sub-pixel mapping result of MAP algorithm, (h) is the sub-pixel mapping result of SPMv algorithm, (i) is the sub-pixel mapping method based on the distribution characteristics of spatial objects used in the present invention. Compared with other inventions, the present invention has more accurate results when dealing with linear features, because the present invention fully considers the connectivity of linear features. Moreover, the present invention fully demonstrates the details of planar features and ensures higher accuracy.

Parts not described in detail in the present invention belong to the known techniques of those skilled in the art.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of the claims of the present invention.

Claims (8)

1., based on a remote sensing sub-pixed mapping drafting method for atural object spatial distribution characteristic, it is characterized in that performing step is as follows:

Step 1, pre-service is carried out to input remote sensing image, and then the component ratio utilizing neural classifier to obtain in remote sensing image each pixel shared by often kind of atural object classification is called abundance or mark image; Described often kind of atural object classification is three kinds of atural objects, i.e. punctual geo-objects, linear ground object and area feature;

Step 2, utilize region growing model, by the zones of different that mark Image Segmentation becomes, each region after adopting ordered data storage organization to store segmentation, is convenient to calculate the area in each region, girth and border;

The type of ground objects of step 3, zoning: the shape density index in the region of mark image after computed segmentation, according to corresponding discrimination threshold, realizes the division of punctual geo-objects, linear ground object and area feature;

Step 4, three kinds of atural object division results according to step 3, sub-pixed mapping drafting method based on linear ground object template matches is adopted to linear ground object mixed pixel, sub-pixed mapping drafting method based on vector border is adopted to area feature mixed pixel, sub-pixed mapping drafting method based on spatial coherence and Model Matching is adopted to punctual geo-objects mixed pixel, obtains punctual geo-objects, linear ground object, area feature sub-pixed mapping charting results;

Step 5, inlay the sub-pixed mapping charting results of punctual geo-objects, linear ground object, area feature, finally obtain sub-pixed mapping charting results.

2. a kind of remote sensing sub-pixed mapping drafting method based on atural object spatial distribution characteristic according to claim 1, is characterized in that: described step 2 utilizes region growing model, and the zones of different specific implementation process become by mark Image Segmentation is as follows:

(1) a random selecting n pixel in mark image, as seed;

(2) region at seed place is expanded, until all pixels of mark image are all divided into specific region, for the neighborhood pixel of seed, if the threshold epsilon that the difference of the pixel value of its pixel value and seed is specifying, then expand this neighborhood pixel in the region at current seed place.

3. a kind of remote sensing sub-pixed mapping drafting method based on atural object spatial distribution characteristic according to claim 2, is characterized in that: in be 1/S, S the be sub-pixed mapping drawing of described threshold epsilon value by low resolution to high-resolution amplification factor.

4. a kind of remote sensing sub-pixed mapping drafting method based on atural object spatial distribution characteristic according to claim 1, it is characterized in that: in described step 3, the shape density index in the region of mark image after computed segmentation, according to corresponding discrimination threshold, the division realizing punctual geo-objects, linear ground object and area feature is implemented as;

I＝ρ ₁×S _shape+ρ ₂×D _density

$S_{shape}=\frac{L}{4\&times;\sqrt{A}},D_{density}=\frac{N}{M}$

Wherein, I represents shape density index, S _shaperepresent with the shape index in region, L represents the pixel number of zone boundary, and A represents the area in region, D _densityrepresent the dnesity index in region, N represents the pixel number of region of out bound rectangle, and M represents the number of region of out bound square boundary pixel, ρ ₁, ρ ₂represent weighted index, wherein ρ ₁+ ρ ₂=1;

If S>=2.3, D≤1.1, I>=1.6, this region is linear ground object; Otherwise this Region dividing is area feature; For remaining region, fractional value is greater than 1/S ²pixel be divided into punctual geo-objects.

5. a kind of remote sensing sub-pixed mapping drafting method based on atural object spatial distribution characteristic according to claim 1, it is characterized in that: described step 4, linear ground object template matching method is adopted to the method for linear ground object sub-pixed mapping drawing, comprise definition linear die, coupling linear die and the sub-pixed mapping of linear ground object to chart three steps, specific as follows:

(1) linear die is defined: utilize the linear die of 20 kind of 3 × 3 pixel to carry out template matches, linear die comprises straight line and curve type;

(2) matching template is linear: utilize the related coefficient of 3 × 3 local windows of 3 × 3 template pixels and mark image to represent their correlativity, computing formula is as follows:

$r_{jk}=\frac{\underset{m=-1}{\overset{1}{\&Sigma;}}\underset{m=-1}{\overset{1}{\&Sigma;}}{T}_k(m,n)\&times;F_{jc}(x+m,y+n)}{\sqrt{\underset{m=-1}{\overset{1}{\&Sigma;}}\underset{m=-1}{\overset{1}{\&Sigma;}}{T}_k{(m,n)}^2\&times;\underset{m=-1}{\overset{1}{\&Sigma;}}\underset{m=-1}{\overset{1}{\&Sigma;}}{F}_{jc}(x+m,y+n)}}$

Wherein, F _jcrepresent a jth pixel of C class atural object mark image, x, y represent the coordinate of this pixel, T _krepresent a kth linear die, m, n represent the position of pixel in 3 × 3 pixel linear die, and specify that the coordinate of top left corner pixel is-1 ,-1, and the present invention adopts and makes r _jkmaximum linear die is as the optimal Template of this mixed pixel; (x+m, y+n) represents the neighborhood pixel of a jth pixel, at the fractional value of C class atural object mark image;

(3) the sub-pixed mapping drawing of Linear feature: for F _jcabove-mentioned two steps determine the best trend of linear ground object in this mixed pixel, the i.e. trend of best linear template, definition pixel j top left co-ordinate is (0,0), lower right corner coordinate (1,1), make the size of linear die equal with mixed pixel j, then in linear die, the coordinate of pixel passes through following formulae discovery:

$\left\{\begin{array}{c}{m}^{\&prime;}=\frac{2\&times;(n+2)-1}{6}\\ n^{\&prime;}=\frac{2\&times;(n+2)-1}{6}\end{array}\right.,m,n=-1,0,1$

The pixel of each sub-pixed mapping in mixed pixel can represent with following formula:

$\left\{\begin{array}{c}{x}^{\&prime;}=\frac{2\&times;x-1}{2\&times;S}\\ y^{\&prime;}=\frac{2\&times;y-1}{2\&times;S}\end{array}\right.,x,y=1,2,...,S$

Wherein, x, y are the ranks number of sub-pixed mapping, and S is the multiple of amplification factor and image augmentation, calculate new pixel i.e. each sub-pixed mapping to optimum linear template T _kintermediate value is the minimum Eustachian distance of 1 pixel, carries out ascending order arrangement to these minimum euclidean distances, front F _cj× S ²the type of ground objects of sub-pixed mapping be C class atural object.

6. a kind of remote sensing sub-pixed mapping drafting method based on atural object spatial distribution characteristic according to claim 1, is characterized in that: the method for described step 4 pair area feature sub-pixed mapping drawing adopts the sub-pixed mapping drafting method based on vector border to comprise following four steps:

(1) for each mixed pixel with vector border extraction model, obtain the length and location of every bar line segment on mixed pixel inner boundary;

(2) clockwise every bar line segment of generating of Connection Step (1) successively, obtains the initial Polygonal Boundary of C class atural object in mixed pixel;

(3) every bar line segment intersection of the initial Polygonal Boundary that obtains of delete step (2) and step (1), namely obtains final vector border;

(4) use ray diagnostic method, carry out assignment to the atural object in vector border.

7. a kind of remote sensing sub-pixed mapping drafting method based on atural object spatial distribution characteristic according to claim 1, is characterized in that: the sub-pixed mapping drafting method based on spatial coherence and Model Matching that described step 4 pair punctual geo-objects adopts, and detailed process is as follows;

(1) to the mark image of C class type of ground objects, area feature sub-pixed mapping chartography is first adopted to obtain the image of its sub-pixed mapping drawing;

(2) to the punctual geo-objects pixel j in the mark image of C class type of ground objects, the not blue exponential quantity I of its place local window is calculated; For the image that this mixed pixel is obtained by sub-pixed mapping drafting method, calculate 3 × 3 × S of mixed pixel j ²not blue index I';

(3) if | I'-I|≤θ, θ distinguish the threshold value whether pixel level punctual geo-objects mate with sub-pixed mapping ground shape atural object space distribution, then the image that this sub-pixed mapping charts is the net result of punctual geo-objects; If | I'-I| >=θ, the property value of two pixels in the sub-pixed mapping drawing image of random adjustment mixed pixel j, recalculating not blue exponential quantity, until meet | I'-I|≤θ, obtains the punctual geo-objects sub-pixed mapping charting results of the mark image of C class type of ground objects.

8. a kind of remote sensing sub-pixed mapping drafting method based on atural object spatial distribution characteristic according to claim 1, it is characterized in that: punctual geo-objects, linear ground object, the integrated process obtained in final sub-pixed mapping charting results of inlaying of area feature sub-pixed mapping charting results are by described step 5: first by the sub-pixed mapping of area feature drawing image layer as a setting, superpose the sub-pixed mapping charting results of punctual geo-objects and the sub-pixed mapping drawing image of linear ground object successively; In the process of inlaying, if the pixel value of sub-pixed mapping drawing image is null, then use 3 × 3 window traversal neighborhoods, the classification that occurrence number is maximum is the category attribute of this pixel.