JP3742882B2 - Polarization synthetic aperture radar image processing method and apparatus - Google Patents

Description

  The present invention relates to a polarization synthetic aperture radar image processing method and apparatus for performing land cover classification from image data obtained by a high resolution polarization synthetic aperture radar mounted on a satellite or an aircraft that observes the earth surface.

  INDUSTRIAL APPLICABILITY The present invention is used in the field of remote sensing using radio waves, particularly in the field relating to data processing of a synthetic aperture radar mounted on a satellite or aircraft that uses polarization information. As an application of land cover classification, it can be used for environment, disaster monitoring, auxiliary data for land use planning, etc.

  Polarization synthetic aperture radar applied to synthetic aperture radar onboard satellites and aircraft has been developed and is now in operation. A synthetic aperture radar device is a device that obtains two-dimensional data for emitting radio waves from a moving platform such as an artificial satellite or an aircraft to the ground on the side in the traveling direction to reproduce a ground image. In this synthetic aperture radar apparatus, synthetic aperture technology is used to improve the azimuth distance resolution. Synthetic aperture technology is a technology that effectively obtains high resolution equivalent to that using an extremely large-diameter antenna by using the movement of the mounting platform (see Patent Document 1).

  Synthetic aperture radar is a type of image radar, and its image (SAR image) is obtained by two-dimensionally mapping scattered power from an object on the ground surface. The scattered power is affected by the roughness of the object, the complex dielectric constant, the shape, the frequency of the transmission signal, the polarization, the incident angle, and the like. That is, more information can be obtained from the target object by changing the frequency, polarization, incident angle, and the like.

  The polarization synthetic aperture radar apparatus is an apparatus that focuses on the polarization among those that affect the scattering power described above and aims to obtain as much information on the target object as possible by changing the polarization. The antenna unit of this polarization synthetic aperture radar apparatus includes a horizontally polarized antenna and a vertically polarized antenna, and transmits horizontally polarized waves and vertically polarized waves. For this transmission, the scattered wave from the object on the ground surface includes a vertical polarization component and a horizontal polarization component, which are received by two antennas.

  That is, the radar transmitting antenna and the receiving antenna each have a function of transmitting and receiving horizontal polarization (H) and vertical polarization (V), so that HH (reception polarization and transmission polarization) polarization transmitted and received by horizontal polarization is transmitted. Input is a polarization signal of a wave signal, a VH signal transmitted with horizontal polarization / received with vertical polarization, a HV signal transmitted with vertical polarization / received with horizontal polarization, and a VV signal transmitted / received with vertical polarization. become.

In this way, in the polarization synthetic aperture radar apparatus, the transmission antenna and the reception antenna of the radar have a function of transmitting and receiving horizontal polarization (H) and vertical polarization (V), respectively. Four data of HH, HV, VH and VV can be obtained. However, since it is known that HV = VH, there are actually three. In the case of a synthetic aperture radar, a combination of data consisting of scattering coefficients of HH, HV, VH, and VV is obtained for each pixel of an image. This data is generally called a scattering matrix. By analyzing this scattering matrix, the land cover situation on the ground can be examined. As an example of a conventional method of analyzing the scattering matrix
(1) Three-component decomposition of the scattering matrix
(2) Polarimetric Entropy
(3) Three-component scattering model
Etc. are known (see Non-Patent Documents 1 to 3). Based on the characteristics obtained from these methods, supervised classification such as maximum likelihood and unsupervised classification such as ISODATA are performed to estimate the land cover situation.

  A disadvantage of the conventional method is that it is difficult to classify urban areas. As shown in FIG. 4, there are surface scattering, double reflection scattering, and volume scattering in the radio wave incident angle (EL) direction. Surface scattering is the primary Bragg scattering process in sea areas, agricultural land, and low vegetation areas, and for the forest area, it mainly represents scattering from the leaf surface. Twice reflection scattering is a scattering process in which the light is incident on the ground surface and reflected by the tree trunk. Volume scattering is a scattering process in which randomly inclined wires are synthesized, and is mainly caused by HV polarization. As shown in the figure, in natural terrain, natural vegetation such as the ocean, grassland, and forest is generally called azimuth symmetry, and the scattering coefficients of surface scattering, double reflection scattering, and volume scattering are azimuth ( AZ) direction independent.

On the other hand, in artificial terrain, it is considered that there is not much influence of volume scattering like the vegetation layer (Vegetation layer) in buildings such as urban areas, etc. Due to the positional relationship between the direction of the building and the radar, The scattering coefficient changes greatly, and cross-polarized light is also generated from twice reflected scattering (see Non-Patent Document 4). For this reason, in urban areas and the like, there is no azimuth symmetry characteristic, so the classification accuracy of these areas may vary greatly depending on the arrangement of buildings and houses in residential areas.

Japanese Unexamined Patent Publication No. 09-178847 E. Krogager, ZH Czyz, "Properties of the sphere, deplane, helix decomposition," Proc. 3rd International Workshop on Radar Polarimetry, vol.1, pp.106-1114,1995 SR Cloude and E. Pottier, "A Entropy Based Classification Scheme for Land Applications of Polarimetric SAR," IEEE Trans. Geosci. Remote Sensing, vol.34, no.2, pp.68-78, March 1996 A. Freeman and SL Durden, "A Three-Component Scattering Model for Polarimetric SAR Data," IEEE Trans. Geosci. Remote Sensing, vol.36, no.3, pp.963-973, May 1998 G. Franceschetti, A. Iodice and D. Riccio, "A Canonical Problem in Electromagnetic Backscattering from Buildings," IEEE Trans. Geosci. Remote Sensing, vol.40, no.8, pp.1787-1801 August 2002 Yoshio Yamaguchi, Basics and Applications of Polarimetric Radar, Realize, 1998

  In Freeman's algorithm (the above-described three-component scattering model decomposition method), a component in which cross polarization occurs is treated as volume scattering. However, in urban areas and residential areas, cross-polarized waves are generated from double reflection scattering depending on the direction of the building. Therefore, in the conventional method, when cross polarization occurs in an urban area, it may be erroneously determined that volume scattering has occurred.

  SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to solve such problems and improve the classification accuracy of urban areas by reducing the erroneous determination by improving the method for determining natural terrain and others.

  The polarization synthetic aperture radar image processing method of the present invention reads and reads polarization complex image data that is a two-dimensional image acquired with each polarization and is given a complex scattering coefficient for each pixel. Between the HH image transmitted and received in the horizontal polarization of the polarization complex image data and the HV image transmitted in the vertical polarization / received in the horizontal polarization and the VH image transmitted in the horizontal polarization / received in the vertical polarization and the vertical polarization. The correlation between the VV image transmitted and received by the wave is calculated for each pixel, and the calculated correlation value of each pixel is separated and determined into a natural terrain and other terrain based on the reference value. The feature amount is calculated with the natural terrain model for the obtained pixel, and the feature amount is calculated with the other model for the pixel determined to be other terrain, and the land cover status is classified from both of the calculated feature amounts. It consists of conduct.

The polarization synthetic aperture radar image processing apparatus of the present invention is an image for reading polarization complex image data that is a two-dimensional image acquired with each polarization and is provided with a complex scattering coefficient for each pixel. Between the data acquisition processing unit and the HH image transmitted / received by the horizontal polarization of the read polarization complex image data and the HV image transmitted / received by the vertical polarization / transmission / vertical polarization by the horizontal polarization A correlation processing unit for calculating for each pixel a correlation between the VH image received in step S4 and a VV image transmitted and received in vertical polarization, and the calculated correlation value of each pixel based on the reference value, A separation processing unit that determines separation into terrain, a natural terrain scattering matrix decomposition processing unit that calculates features for natural terrain using a natural terrain model, and other terrain pixels. Artificial topography for a scattering matrix decomposing unit for calculating a feature quantity in the other models, the land cover classification processing unit for classifying the land cover status from the feature of both the calculated consists.

According to the present invention, it is possible to reduce misjudgments and improve the classification accuracy of urban areas by improving natural terrain and other judging methods. According to the present invention, it is possible to extract features from a polarization synthetic aperture radar image in consideration of the scattering mechanism in the urban area without being significantly affected by the positional relationship between the urban structure and the radar in the azimuth direction. Become.

  In general, in a high-resolution polarization synthetic aperture radar, a combination of transmission and reception antennas, for example, a horizontal polarization antenna is called H and a vertical polarization antenna is called V, so that an image of a combination of HH, HV, VH, and VV can be transmitted and received. Obtainable. A complex scattering coefficient of each polarization is given for each pixel of the image, and the scattering matrix S can be configured by combining the scattering coefficients.

  In general, when the transmitting and receiving antennas of the radar are at the same position, the elements that are non-diagonal terms of the scattering matrix are equal (SHV = SVH), and the scattering matrix can be expressed by three variables. The covariance matrix of each variable of the scattering matrix is called a Covariance Matrix.

Here, <> means an average between pixels (ensemble average). Generally, it is known that <SHHSHV *>, <SVHSVV *> = 0 holds for natural topography such as ocean, grassland, and forest. This means that there is no correlation between the like polarization (HH, VV polarization) and the cross polarization (HV, VH polarization). Here, * represents a complex conjugate.

  The present invention is an improvement of Freeman's three-component scattering model in the prior art so that it can be applied to urban areas. As shown in (Equation 4) and (Equation 5), the method referred to by this method is to scatter the natural terrain such as ocean, grassland, forest, etc. by surface scattering (Ssurface), double reflection scattering (Sdihedral), volume scattering ( Svolume) is modeled as a composite. On the other hand, in the urban area, surface scattering (Ssurface), double reflection scattering (Sdihedral), improved double reflection scattering (S'dihedral), wire scattering (Swire), and the like are mainly generated. So we extended Freeman's method to model urban scattering. Specifically, the model was divided into natural terrain and others. In addition, the model other than the natural topography is improved so that volume scattering does not occur and the generation of cross polarization can be expressed by double reflection scattering.

  In vegetation that becomes azimuth symmetry, it is said that there is no correlation between like polarization (HH, VV polarization) and cross polarization (HV, VH polarization). This feature can be used to create a criterion for separating natural terrain from other structures.

  In addition, the present invention can handle the case where cross polarization occurs by extending the reflection model twice based on Freeman's method. Furthermore, it is possible to handle scattering of wires and the like from the extended two-reflection model.

  FIG. 1 is a diagram illustrating a polarization synthesis aperture radar image processing apparatus according to the present invention. The “image data acquisition processing unit” 1 is instructed on the polarization complex image data to be processed, and the image data is read. The polarization complex image data is a two-dimensional image of HH, HV, VH, and VV polarization acquired for each polarization, and a complex scattering coefficient is given for each pixel.

  The “HH image and HV image, VH image and VV image correlation processing unit” 2 examines the correlation between images using the following (Equation 6) and (Equation 7). The calculation is performed by the following formula using MxN data centering on the coordinates (i, j) of the image of interest.

これを各ピクセルに対して行う。

This is done for each pixel.

  The “artificial landform and natural landform separation processing unit” 3 performs the following determination on the calculated correlation values Cor (SHH, SHV *) and Cor (SVH, SVV *) of each pixel. If Cor (SHH, SHV *) or Cor (SVH, SVV *) is greater than the reference value, it is determined that the area is other than natural terrain (artificial terrain). When it is other than the above judgment, it is judged as natural terrain.

  The “Natural Terrain Covariance Matrix Decomposition Processing Unit” 4 calculates a feature amount of a pixel determined to be natural terrain using a natural terrain model.

  The “Covariance Matrix Decomposition Processing Unit for Artificial Terrain” 5 calculates the feature amount of the pixel determined to be other (artificial landform) using the other (artificial landform) model.

  The “land cover classification processing unit” 6 classifies the land cover status from the two feature quantities obtained above by using a database with a maximum likelihood method of supervised classification or an ISODATA method of unsupervised classification. Then, the classified result is displayed.

  The “natural terrain covariance matrix decomposition processing unit” 4 shown in FIG. 1 will be described in more detail. The pixel determined to be natural terrain in the “artificial terrain and natural terrain separation processing unit” 3 can decompose the scattering matrix based on the method described in “Non-Patent Document 3”, and can be applied to classification. The amount can be calculated. In this method, as shown in FIG. 2, the natural terrain is decomposed as a result of surface scattering, double reflection scattering, and volume scattering. Further, it is assumed that cross polarization occurs only from volume scattering. Therefore, it is not suitable for urban areas. In the actual calculation, the above model is modeled as follows.

1) Surface scattering model:
As shown in FIG. 2A, the surface scattering is a primary Bragg scattering process in a sea area, agricultural land, and a low vegetation area. For the forest area, it mainly represents scattering from the leaf surface. Each element of the scattering matrix and the Covariance Matrix can be expressed as follows.

2) Twice reflection scattering model:
As shown in FIG. 2B, the double reflection scattering is a scattering process that is incident on the ground surface and reflected by the tree trunk, or vice versa. Each element of the scattering matrix and the Covariance Matrix can be expressed as follows. Where the parameter Rpq related to α represents the reflection coefficient of Fresnel, p represents the polarization (v: vertical polarization, h: horizontal polarization), and q represents the reflection location (g: ground surface, t: trunk) . Γp represents a change in phase and attenuation due to polarization.

3) Volume scattering model:
Volume scattering is a scattering process in which randomly inclined wires are synthesized as shown in FIG. This is mainly due to HV polarization. Each element of the Covariance Matrix can be expressed as follows:

  Since the measurement data is a composite of three models, each element of the Covariance Matrix can be described as follows.

  fs, fd, and fv represent the contributions of surface scattering, double reflection scattering, and volume scattering, respectively. α and β are coefficients obtained from components of twice reflection scattering and surface scattering. fv can be derived directly from <ShvShv *>. The remaining parameters are derived from the sign of the real part of <SHHSVV *>

The other parameters fs, fd, α, and β are estimated.
From the above, unknown parameters for the three models can be obtained. Then, a feature amount used for classification is calculated from the obtained parameters. The feature value is the power contribution Ps such as surface scattering, double reflection scattering, and volume scattering for the total P of received power <SHHSHH *>, <SHVSHV *>, <SVHSVH *>, <SVVSVV *> between each polarization, Pd and Pv. The power contributions Ps, Pv, and Pd such as the total power P of each polarization, surface scattering, double reflection scattering, and volume scattering are given as follows.

  The "Covariance Matrix decomposition processing unit for artificial terrain" 5 shown in FIG. 1 will be described more specifically. In the “artificial terrain / natural terrain separation processing unit” 3, the pixel determined to be other (artificial terrain) decomposes the scattering matrix by the following method and calculates a feature quantity applicable to classification. The following points are considered as conditions.

(1) As shown in FIG. 3, scattering on artificial terrain mainly consists of surface scattering, double reflection scattering, improved double reflection scattering, and wire scattering.
(2) The condition of <SHHSHV *>, <SVHSVV *> = 0 does not hold in this landform. However, in each model of surface scattering and twice reflection scattering of Example 2, the cross polarization component is zero, so <SHHSHV *>, <SVHSVV *> = 0. On the other hand, Non-Patent Document 4 shows that in an artificial structure such as a building, a cross polarization component is generated from twice reflected scattering. Therefore, the two-reflection scattering model is extended so that the generation of the cross polarization component can be considered. As a result, the condition of <SHHSHV *>, <SVHSVV *> ≠ 0 can be satisfied from the extended double reflection scattering. Even in a single wire scattering model, <SHHSHV *>, <SVHSVV *> ≠ 0 can be satisfied. In the actual calculation, the above model is modeled as follows.

1) Surface scattering model (see FIG. 3A):
Each element of the scattering matrix and the Covariance Matrix of the surface scattering model can be expressed as follows.

2) Double reflection scattering model (see FIG. 3B):
Each element of the scattering matrix and the Covariance Matrix of the double reflection scattering model can be expressed as follows.

3) Improved double reflection scattering model (see FIG. 3C):
The elements of the scattering matrix and the Covariance Matrix of the improved two-reflection scattering model can be expressed as follows.

4) Wire scattering model (see FIG. 3D):
Each element of the scattering matrix of the wire scattering model (see Non-Patent Document 5) and the Covariance Matrix can be expressed as follows.

Here, since the models 3) and 4) have the same expression format, the measurement data is considered to be a composite of three types 1) and 2), 3) or 4), and each Covariance Matrix element is described as follows: To do.

<SHHSHV*>、<SHVSVV*>からα´が算出できることから、

Since α ′ can be calculated from <SHHSHV *> and <SHVSVV *>,

And parameters ρ and fw or fd ′ can be obtained. Further, the remaining parameters fs, fd, α, and β can be obtained in the same manner as the method of the second embodiment.

From the above, it is possible to obtain unknown parameters for models of surface scattering, double reflection scattering, extended double reflection scattering, and wire scattering. Then, a feature amount used for classification is calculated from the obtained parameters. The feature amount is the power contribution Ps such as surface scattering, double reflection scattering, wire scattering, etc. for the total power of received power <SHHSHH *>, <SHVSHV *>, <SVHSVH *>, <SVVSVV *> between each polarization. Pd, Pw. However, the contribution of the extended double reflection scattering model is calculated as a part of the conventional double reflection scattering contribution. The power contributions Ps, Pd, and Pw, such as the total power P of each polarization, surface scattering, double reflection scattering, and wire scattering, are given as follows.

It is a figure which illustrates the polarization synthetic aperture radar image processing apparatus of this invention. It is a figure which shows the concept of the surface scattering in natural landform, twice reflection scattering, and volume scattering. It is a figure which shows the concept of the surface scattering in an artificial landform, 2 times reflection scattering, the improved 2 times reflection scattering, and wire scattering. It is a figure explaining scattering in natural terrain and artificial terrain.

Explanation of symbols

1 Image data acquisition processing unit 2 Correlation processing unit between HH image and HV image, VH image and VV image 3 Artificial landform and natural landform separation processing unit 4 Covariance Matrix decomposition processing unit for natural landform 5 Covariance Matrix decomposition processing for artificial landform Part 6 Land cover classification processing part

Claims (2)

Read the polarization complex image data that is a two-dimensional image acquired with each polarization and is given a complex scattering coefficient for each pixel,
Between the HH image transmitted and received in the horizontal polarization of the read polarization complex image data and the HV image transmitted in the vertical polarization / received in the horizontal polarization and the VH image transmitted in the horizontal polarization / received in the vertical polarization. And the correlation between the VV image transmitted and received with vertical polarization for each pixel,
Based on the calculated correlation value for each pixel based on the reference value, the natural terrain and other terrain are separated and determined.
Calculate features for natural terrain pixels using natural terrain models, and calculate features for other terrain pixels using other models,
The land cover status is classified from both of the calculated features.
A polarization synthetic aperture radar image processing method comprising: An image data acquisition processing unit for reading polarization complex image data that is a two-dimensional image acquired at each polarization and is provided with a complex scattering coefficient for each pixel;
Between the HH image transmitted and received in the horizontal polarization of the read polarization complex image data and the HV image transmitted in the vertical polarization / received in the horizontal polarization and the VH image transmitted in the horizontal polarization / received in the vertical polarization. And a correlation processing unit that calculates a correlation between each pixel and a VV image transmitted / received by vertical polarization,
Based on the reference value of the calculated correlation value of each pixel, a separation processing unit that separates the natural terrain from the other terrain,
A scattering matrix decomposition processing unit for natural terrain that calculates a feature amount using a natural terrain model for pixels determined to be natural terrain,
A scattering matrix decomposition processing unit for artificial terrain that calculates a feature amount using another model for pixels determined to be other terrain,
A land cover classification processing unit for classifying the land cover status from both of the calculated feature amounts;
Polarization synthetic aperture radar image processing apparatus comprising:

Images