CN103106655B - The non-supervisory extracting method in a kind of building site based on remote sensing image - Google Patents

Abstract

For the feature of building site, invent the non-supervisory extraction method in a kind of construction site based on remote sensing image.The method utilizes the texture features of building site in remote sensing image to regenerate seed points, is formed at by growth and builds region, it is achieved the non-supervisory extraction in construction site.Main point of four steps: (1) removes the green vegetation having similar grain feature in image with construction site；(2) image is transformed into YCbCr space；(3) calculate each pixel and its similarity of pixel in 3 × 3 neighborhoods, obtain standby seed points, calculate the Euler's distance between each standby seed points and its neighbours' pixel, it is judged that obtain seed points；(4) according to certain rule, the seed points of close quarters is carried out growth to be formed at and build region.Invention has been filled up currently for the disappearance at building constructions thing extraction algorithm, provides strong foundation for investment construction progress monitoring, casualty loss detection and government decision, is identified having important function for having the target of particular texture feature.

Description

The non-supervisory extracting method in a kind of building site based on remote sensing image

Technical field:

The present invention is a kind of building site practical, efficient recognition methods, it is adaptable to high resolution remote sensing satellite data and unmanned plane are taken photo by plane data, are specifically related to pattern recognition and digital image processing techniques.Can extract in construction site, urban development monitoring, the engineering construction progress aspect such as monitoring, disaster detection, assessment have a wide range of applications.

Background technology:

Building extracts one of basic problem being by monitoring of land use, provides reliable foundation for urban development monitoring analysis, investment construction process monitoring and government decision.

The building that the algorithm being identified building currently with remote sensing image is both for being currently in use extracts, and have ignored the building built.These methods are based primarily upon the information such as spectral information and texture, shape, edge, according to shape, height and color characteristic that building is basic, extract, by setting empirical value and establishment model, the building built up.Can only be and higher to their spectral information and the feature of texture information and distribution restriction ratio for having given shape or spectral information and the building of height, to texture and spectral information more complicated just at the identification blank out especially of building constructions thing.

Present invention utilizes region growing thought, propose a kind of building site extracting method, can be identified at the building built.The formation condition of the spectral characteristic setting seed points that application and construction building site is special, extracts the seed points being in spectral signature change intense regions, and is classified as the point in its contiguous range in building site by formulating new growing strategy.

Summary of the invention:

At present, building recognition is primarily directed to the building built well, does not includes the construction site built.And the method commonly used is according to the major part distinctive spectrum of building and the information design such as textural characteristics and geometry, there is the biggest randomness, being only applicable to some specific environment, only specific building is had good extraction effect, universality is the highest.

The present invention is a kind of non-supervisory construction site extraction method, for the feature that construction site spectrum and texture information are trifling, selected spectrum and texture concentrate the seed points of region of variation, give the growing strategy that region growing is new, utilize the method for region growing to be extracted accurately and efficiently in construction site.

Concrete method step is as follows:

The first step: remove unmanned plane aviation image Green vegetation

Green vegetation occupies the biggest ratio in remote sensing images, and they have the texture information that change is violent as the building built, and can extract construction site and interfere.Owing to their spectral information is more stable, consistent, it is possible to use first vegetation is removed by analysis of spectrum threshold.

Second step: color of image space is transformed into YCbCr from RGB

Between RGB color three passages R, G and B, there is the highest dependency, be suitable for showing image.But the distance between pixel in rgb space can not express mankind's difference to different atural object perception on consistent yardstick, is therefore not suitable for utilizing rgb space to be analyzed spectrum and the texture of atural object.(Y refers to luminance component to YCbCr, Cb refers to chroma blue component, and Cr refers to red chrominance component) it is applicable to image segmented extraction, and the perception of color distortion can be represented by Euler's distance and (sees document: Shih by the mankind, F.Y.andS.Cheng, Automaticseededregiongrowingforcolorimagesegmentation.Im ageandVisionComputing, 2005.23 (10): p.877-886).Therefore we utilize YCbCr color space to be analyzed aviation image atural object.

3rd step: automatically generate seed points

1, the similarity of each pixel and pixel about is calculated

Setting window size as 3 × 3, image variance on a certain wave band in window ranges is:

${\mathrm{\σ}}_x=\sqrt{\frac{1}{9}\underset{i=1}{\overset{9}{\Σ}}{(x_i-\stackrel{\‾}{x})}^2}---\left(1\right)$

Wherein, x is tri-passages of Y, Cb and Cr.

$\stackrel{\‾}{x}=\frac{1}{9}\underset{i=1}{\overset{9}{\Σ}}{x}_i---\left(2\right)$

Population variance can be expressed as:

σ=σ_Y+σ_Cb+σ_Cr(3)

It is standardized as:

${\mathrm{\σ}}_N=\frac{\mathrm{\σ}}{{\mathrm{\σ}}_{max}}---\left(4\right)$

Wherein, σ_maxIt it is variance maximum in 3 × 3 windows with all pixels of view picture image as core

So, the spectral similarity of 3 × 3 windows with a certain pixel as core may be defined as:

H=1-σ_N(5)

If calculated similarity is more than some threshold value, then this pixel is classified as seed standby point.This threshold value is to be determined by ' Otsu ' method.' Otsu ' method is also difference method between maximum kind, and its threshold value determined can divide the image into foreground and background two class, and makes inter-class variance maximum, and variance within clusters is minimum.Core concept is:

T=Max [w₀(t)×(u₀(t)-u)²+w₁(t)×(u₁(t)-u)²](6)

Wherein t is the threshold value of segmentation, w₀(t) be threshold value be the ratio shared by backdrop pels number during t, w₁(t) be threshold value be the ratio shared by prospect pixel number during t, u₀(t) be threshold value be background mean value during t, correspondingly, u₁T () is prospect average.U is the average of entire image, it may be assumed that

U=w₀×u₀+w₁×u₁(7)

In the computing formula of threshold value, w₀(t)×(u₀(t)-u)²+w₁(t)×(u₁(t)-u)²It is the calculating variance between foreground and background two class.The threshold value i.e. making inter-class variance maximum is ' image segmentation threshold that determines of Otsu ' method.

2, calculate standby seed points in 3 × 3 windows, be adjacent Euler's distance corresponding between pixel

Wherein, Euler's distance can be defined as:

$d_i=\frac{\sqrt{{(Y-Y_i)}^2+{(C_b-C_{b_i})}^2+{(C_r-C_{r_i})}^2}}{\sqrt{Y^2+C_b^2+C_r^2}}---\left(8\right)$

I=1,2 ..., 8

And Euclidean distance maximum between pixel and its neighbours' pixel is:

$d_{max}=\underset{i=1}{\overset{8}{max}}\left(d_i\right)---\left(9\right)$

In standby seed points, if the Euclidean distance of certain point maximum is more than 0.05, then it is set as seed points.

4th step: region increases

Traditional region growing method is to divide the image into as multiple regions, and becomes to have the true atural object of certain textural characteristics and smooth spectral information by region merging technique.But for the building built, owing to having the strongest spectrum brokenness and texture discordance, it is difficult with traditional method and region is merged.Therefore we are according to the formation feature of seed points, have invented a kind of new region growing method.

1, judge whether seed points is in close quarters

The place that seed points owing to generating is in spectrum and texture information all changes greatly, and comparatively dense is compared in the spectrum of the building in process of construction and texture information change.Therefore, there is intensive seed points in region, construction site.In order to judge whether seed points belongs to high-density region, we are centered by each seed points, add up the number of seed points in 5 × 5 window ranges.If it exceeds 16, then it is assumed that it is in high density area.

The size of the window ranges of statistics seed points number determines the seed points sensitivity to closeness.If window ranges is the least, then it is extremely difficult to the standard of closeness, the seed points being much positioned at region, construction site so can be caused to be lost, make construction site be identified as vacant lot by wrong.If window ranges is very big, can easily reach closeness standard, can increase the construction site area detected, cause erroneous judgement, the building making vacant lot be taken as in process of construction extracts.Through it is demonstrated experimentally that 5 × 5 windows are more suitable, using 16 as weighing the extraction result of density standard just also in accordance with ground truth.

2, the bounding rectangles of seed point set in close quarters is determined

The just building in process of construction has intensive, discrete seed points set, but is not connected region.In order to by discrete point set compartmentalization, we are in highdensity seed point set, centered by each seed, in 5 × 5 regional extents, find the boundary rectangle of the set that all seed points constitute, and the non-seed point of rectangle inside is collectively labeled as seed points.So, discrete seed point set grows into as continuous, smooth region, is the extraction result in construction site.

Accompanying drawing illustrates:

Fig. 1 is the flow chart of the present invention；

Fig. 2 is the explanation figure of the new Rule of Region-growth formulated；

Fig. 3 is for specifically emulating case；

Detailed description of the invention:

Fig. 2 is the explanation of the region growing principle that the present invention proposes.Its Green point is non-seed point, and yellow is selected seed points, the red core seeds point for processing.In this example, in 5 × 5 neighborhoods, there are 17 seed points at red o'clock, more than the threshold value 16 of the judgement dense degree that the present invention sets, therefore after region growing, the seed points that is the most all marked as in the range of 5 × 5, the building site i.e. monitored.

Fig. 3 is the process using this method to extract the construction site of high-resolution unmanned plane aviation image

Test intercepts the unmanned plane aviation image shot Yunnan Airport on May 30th, 2009, and size is: 886x554

Fig. 3 (a) is Yunnan Airport part basis construction area figure.In figure, atural object is based on construction area, wherein has part more scattered (see atural object in blue circle), and major part compares concentration.Subregion is also had to belong to non-building area, atural object in show chromosphere.

Fig. 3 (b) is the seed point diagram extracted.Change it will be seen that the closeness of seed points is as construction site compared with intensity.In area, non-construction site, seed points distribution is the most sparse, and the central area intensive in construction site, seed points distribution is the most intensive.

Fig. 3 (c) is the extraction result figure of construction area.What distribution was concentrated is just well extracted in building site, and is distributed around sparse small-sized building site and is also extracted.Non-construction area is also well neglected.Extracting results contrast smooth, complete, geometry is also close with actual atural object, has certain application potential.

Claims (2)

1. construction site based on high-resolution remote sensing image extracting method, it is characterized in that utilizing the formation feature of seed points, according to construction site seed points densely distributed degree, construction site is positioned, discrete seed point set is generated connected region by recycling region growing method, and then generate building region, building site, specific embodiments is as follows:

(1) analysis of spectrum threshold is utilized to be rejected by image Green vegetation

Although green vegetation has more continuous and smooth spectral information, but its texture information is more trifling, and easily the extraction to construction site interferes, and is therefore removed by green vegetation first with spectrum threshold,

(2) image is transformed into YCbCr space from RGB color

The display of remote sensing images uses rgb space more, but owing to the dependency between R, G and B is higher, mankind's difference to different atural object perception cannot be expressed with spectrum intervals, thus cannot be used for image segmentation, target identification processing, and the Y in YCbCr space, Cb and Cr is luminance component respectively, chroma blue component and red chrominance component, more sensitive to spectrum and texture information, it is used for image segmentation, extracts feature of interest, it is thus desirable to by color space conversion

(3) automatic decision and generation seed points

First calculate each pixel and be adjacent the spectral similarity of pixel, and utilize otsu method to automatically generate threshold value, as the evaluation criterion of spectral similarity, if greater than this threshold value, then as the alternative point of seed, then calculate the texture paging between itself and adjacent element, generate seed point set

(4) region growing, generates construction site and extracts result

Texture information and the feature of spectral information for construction site, and the distribution characteristics of seed points, with each seed points as core, judge its seed points number in the range of 5 × 5, if greater than 16, then set it and be belonging to compact district, then all being collectively labeled as in its 5 × 5 neighborhood rectangle is built building site recognition result.

Construction site the most according to claim 1 extracting method, it is characterised in that spectrum and the texture information in construction site are more trifling, match with seed points formation condition, the two are combined.