Image analysis device and image analysis method

Abstract

An image analysis device comprises a signal separation unit 10 that: receives, as input, a plurality of radar images in which the same geographical region is recorded and a phase gradient model that represents the phase differences between a plurality of nearby pixels at the surface of an analysis target that could be present in the radar images; and extracts a component that conforms to the phase differences represented by the phase gradient model from the radar images.

Description

[Technical Field]

[0001] The present invention relates to an image analysis device and an image analysis method for analyzing displacements and the like based on radar images. [Background technology]

[0002] There is a change detection technology that detects areas where the state of the earth's surface has changed based on images taken from a high altitude, such as images taken by an artificial satellite.

[0003] Synthetic Aperture Radar (SAR) technology is a technology in which a radar mounted on a flying object such as a satellite or aircraft transmits and receives electromagnetic waves while the flying object is moving, and obtains images (hereinafter referred to as SAR images) equivalent to images obtained by an antenna with a large aperture. Synthetic aperture radar is used, for example, to analyze ground displacements by processing the signals of waves reflected from the ground surface. Note that the ground surface includes not only the ground but also the surface (top) on which low structures such as low-rise buildings exist.

[0004] An image taken by a flying object such as a satellite is called a radar image. An SAR image is an example of a radar image. Hereinafter, the flying object that transmits and receives electromagnetic waves will be referred to as a satellite, but the flying object is not limited to a satellite.

[0005] Interferometric SAR analysis is an effective method for analyzing ground surface displacement, etc. In interferometric SAR analysis, the phase difference between the radio signals that make up multiple (e.g., two) SAR images taken at different times is calculated. Then, based on the phase difference, the change in distance between the flying object and the reflector that occurred between the times of taking the images is detected.

[0006] SAR images can be affected by layover (see, for example, Patent Document 1). For example, when a power transmission tower or a high-rise building is present in the observation area, the reflected waves from the power transmission tower or the high-rise building are mixed with the reflected waves from the ground in the radar mounted on the artificial satellite. As a result, the waveform of the received waves by the radar mounted on the artificial satellite becomes unstable. Specifically, a layover can occur in SAR images, where information from the power transmission tower overlaps with information from the ground. In a situation where a layover occurs, changes may occur between multiple SAR images even if no changes actually occur in the observation area.

[0007] Taking a power transmission tower as an example, the reflection strength of radio waves from the tower is often weaker than that from the ground. In other words, the power transmission tower is a reflector that exhibits weak reflection. Therefore, the reflected waves from the tower are mixed in with the reflected waves from the ground. As a result, when analyzing a power transmission tower using SAR images, layover makes it difficult to perform an analysis based on the phase of the power transmission tower (including information on the distance from the radar).

[0008] Furthermore, displacement analysis using SAR images generally targets linear displacement, making it difficult to estimate nonlinear displacement of buildings caused by factors such as scouring due to floods or earthquakes (such as large tilts or collapses over a short period of time). [Prior art documents] [Patent documents]

[0009] [Patent Document 1] International Publication No. 2015 / 151134 [Non-patent literature]

[0010] [Non-Patent Document 1] Ricardo Lanari, et al., "A Small-Baseline Approach for Investigating Deformations on Full-Resolution Differential SAR Interferograms", IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 42, NO. 7, pp. 1377-1386, JULY 2004 [Non-patent document 2] Fabrizio Lombardini, et al., "New developments of 4D+ differential SAR tomography to probe complex dynamic scenes", IEEE IGARSS, pp. 3362-3365, 2014 Summary of the Invention [Problem to be solved by the invention]

[0011] SBAS (Small Baseline Subset) is a time-series analysis method using SAR images (see, for example, Non-Patent Document 1). SBAS is a method for estimating displacement using multiple SAR interferograms with short vertical baselines and short imaging time intervals. SBAS is used to convert (unwrap) the relative phase (a phase that takes values between -π and π) into an absolute phase, and then displacement is extracted by filtering the absolute phase using a regression model or the like. Such a method can also handle nonlinear displacement. However, when using such a method, it is only possible to estimate the displacement of a specific reflector. In other words, when both reflectors exhibiting weak and strong reflections exist, it is difficult to evaluate the phase of only the reflector exhibiting weak reflection.

[0012] There is also a method of modeling in advance the phase change of each individual reflector over time each time imaging is performed, using parameters such as displacement speed and height. For example, when using 4D tomography as described in Non-Patent Document 2, a time-varying phase change model is used. The phase change model is created using 4D tomography and includes information about height. The actual observation signal (observation image) is compared with multiple pre-modeled phase change models. Then, a phase change model for each reflector that matches the observation signal is selected. In other words, a model is selected in which the difference between the observation signal, which is a complex image, and the complex number derived from the phase change model is equal to or less than a predetermined value. Desired parameters are obtained from each selected phase change model.

[0013] When such a method is used, it is possible to obtain the displacement of each of multiple reflectors. In other words, even if a layover occurs, it is possible to separate multiple reflectors (e.g., a transmission tower and the ground) and obtain the displacement of each. However, if a phase change model that matches the observed signal does not exist, it is not possible to separate multiple reflectors or estimate their parameters. In addition, since a linear model is generally prepared as the phase change model for a reflector, methods using 4D tomography are not suitable for estimating the nonlinear displacement of a specific reflector.

[0014] In addition, as a method of using multiple models obtained in advance through modeling, there is also a method of using 3D images created using the MUSIC (Multiple Signal Classification) method or the Capon method, and there is also a method of using sparsity of the observed signals (observed images) in combination.

[0015] Whichever known method is used, The heightIt is difficult to separate the waves reflected from high reflectors from those reflected from the ground, estimate their respective displacements (especially those of reflectors such as transmission towers), and also to deal with the nonlinear displacement of the reflectors. For example, even when using 4D tomography, it is difficult to create an appropriate nonlinear model.

[0016] The present invention aims to provide an image analysis device and an image analysis method that can analyze the phase of a reflector exhibiting weak reflection even when a reflector exhibiting weak reflection and a reflector exhibiting strong reflection are observed overlapping each other due to layover, and that can also handle nonlinear displacement of the reflector. [Means for solving the problem]

[0017] The image analysis device according to the present invention includes a signal separation means for receiving a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar image, and extracting from the radar image a component that matches the phase difference represented by the phase gradient model. The signal separation means includes a division means for dividing each of the plurality of radar images into small regions, a phase gradient removal means for removing from each small region a component that matches the phase difference represented by the phase gradient model, and a signal extraction means for extracting a dominant component in the radar image from which the phase difference has been removed. .

[0018] The image analysis method according to the present invention inputs a plurality of radar images of the same area and a phase gradient model representing the phase difference between adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and extracts from the radar images components that match the phase difference represented by the phase gradient model. Dividing each of the plurality of radar images into small regions, removing components that match the phase difference represented by the phase gradient model from each small region, and extracting dominant components in the radar image from which the phase difference has been removed. .

[0019] The image analysis program according to the present invention inputs a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and causes a computer to execute a process of extracting components that match the phase difference represented by the phase gradient model from the radar images. Dividing each of the plurality of radar images into small regions, removing from each small region components that match the phase difference represented by the phase gradient model, and extracting the dominant component in the radar image from which the phase difference has been removed. . [Effects of the Invention]

[0020] According to the present invention, even when a reflector exhibiting a weak reflection and a reflector exhibiting a strong reflection are observed overlapping each other due to layover, the phase of the reflector exhibiting a weak reflection can be analyzed, and further, nonlinear displacement of the reflector can also be handled. [Brief explanation of the drawings]

[0021] [Figure 1] 1 is a block diagram showing an overview of an embodiment of an image analysis device. [Figure 2] 1 is a block diagram illustrating an example of the configuration of an image analysis device according to a first embodiment. [Figure 3] 4 is a flowchart showing the operation of the image analysis device of the first embodiment. [Figure 4] FIG. 10 is a block diagram illustrating an example of the configuration of an image analysis device according to a second embodiment. [Figure 5] FIG. 10 is an explanatory diagram for explaining reflection of an analysis target and reflection of a surrounding area. [Figure 6] 10 is a flowchart showing the operation of the image analysis device of the second embodiment. [Figure 7] FIG. 10 is a block diagram illustrating an example of the configuration of an image analysis device according to a third embodiment. [Figure 8] FIG. 10 is an explanatory diagram for explaining a layover area continuity model. [Figure 9] 10 is a flowchart showing the operation of the image analysis device of the third embodiment. [Figure 10] FIG. 10 is a block diagram showing an example of the configuration of an image analysis device according to a fourth embodiment. [Figure 11] 10 is a flowchart showing the operation of the image analysis device of the fourth embodiment. [Figure 12] FIG. 1 is a block diagram illustrating an example of a computer having a CPU. [Figure 13] FIG. 2 is a block diagram showing the main parts of the image analysis device. DETAILED DESCRIPTION OF THE INVENTION

[0022] Hereinafter, embodiments of the present invention will be described with reference to the drawings. In each of the embodiments described below, an SAR image is used as an example of a radar image obtained using electromagnetic waves, but the radar image is not limited to an SAR image. In addition, in each of the embodiments described below, a power transmission tower is used as an example of a reflector that exhibits weak reflection, but reflectors that exhibit weak reflection are not limited to power transmission towers.

[0023] Each embodiment described below uses a spatial phase gradient model related to a transmission tower. The spatial phase gradient model is a model that constrains the phase gradient between multiple adjacent pixels in an SAR image. Note that the phase gradient constraint means, for example, specifying the degree of phase difference. A signal separation unit, which will be described later, determines that a portion of the SAR image that has a phase gradient that matches the constraint (condition) is a surface belonging to a reflector (transmission tower) that exhibits weak reflection. In other words, the signal separation unit has a function of extracting from the SAR image a component that has a phase gradient that matches the spatial phase gradient model.

[0024] 1 is a block diagram showing an outline of an embodiment of an image analysis device. The image analysis device includes a signal separation unit 10, an SAR image storage unit 20, and a phase gradient model storage unit 30.

[0025] The SAR image storage unit 20 stores N SAR images (for example, N=10 to 30) of images of the same area. Hereinafter, the SAR images stored in the SAR image storage unit 20 are also referred to as input SAR images or a group of input SAR images. The N SAR images are radar images of the same area, and are radar images acquired at different times or on different trajectories. That is, the SAR images stored in the SAR image storage unit 20 may be radar images acquired at different times but on the same trajectory. Furthermore, the multiple SAR images may be radar images acquired at the same time but on different trajectories. Furthermore, the multiple SAR images may be radar images acquired at different times and on different trajectories.

[0026] The input SAR image may be an image after flattening to remove a phase dependent on a known elevation model. The elevation model used in flattening may be a planar or spherical model.

[0027] The phase gradient model storage unit 30 stores a phase gradient model relating to a reflector (transmission tower) that has been created in advance. The signal separation unit 10 extracts signals that match the phase gradient model from each SAR image.

[0028] A phase gradient model is a model that reflects, for example, the three-dimensional shape of the object to be analyzed, the degree of thermal expansion that may occur, the degree of bending, etc., and defines the degree of phase difference that multiple adjacent points on the surface of the object to be analyzed may have on an SAR image. The phase gradient model can be expressed, for example, as a probability model, an evaluation function, or a template array. A template array is sometimes called a template image.

[0029] The phase gradient model may be created by any method, but for example, the phase gradient model can be created by the following method.

[0030] [Method 1: Create a probability density function (phase difference function) that constrains the phase difference between adjacent pixels] For example, the phase of pixel p (the phase to be observed at pixel p) is expressed as θ p , the phase of the adjacent pixel (p+1) (the phase to be observed at pixel p+1) is θ p+1 The complex numbers with absolute value 1 corresponding to these phases are denoted as s p =exp(jθ p ), s p+1 =exp(jθ p+1 ) In the phase gradient model, the phase θ p and phase θ p+1 For example, if pixel p and pixel (p+1) exist in the object of analysis (reflector), a phase gradient model is created that has a phase difference that occurs with a probability proportional to the function shown in the following equation (1):

[0031]

number

[0032] In equation (1), k p denotes a parameter that controls the phase gradient (corresponding to the phase difference) and the strength with which the phase difference between two pixels is constrained. g is a function that indicates an increasing tendency, for example, an exponential function.

[0033] Equation (1) is k p =|k p |exp(j∠k p ), the phase difference (θ p -θ p+1 ) is the phase angle ∠k p This corresponds to a model in which the value is close to |k p | is large, the phase difference (θ p -θ p+1 ) and phase angle ∠k p The difference between |k p | is small, the phase difference (θ p -θ p+1 ) and phase angle ∠k p It also indicates that there is a high possibility that the difference between |k p |=0 indicates that the phase difference is not constrained. p |=∽ indicates that the phase difference is constrained to be equal to the phase angle.

[0034] In addition, k p may be different for each SAR image. For example, when an analysis target having a three-dimensional height is photographed from different orbits, k p also varies depending on the orbit in which the SAR images were taken.

[0035] When the height of the object to be analyzed is significant (has a height equal to or greater than a predetermined value), for example, the topological gradient that occurs according to the gradient of the three-dimensional height can be used as the topological gradient of the model.

[0036] It is known that the phase contained in SAR images taken from a flying object has a phase that depends on the orbit and the height of the object being photographed. For example, among the orbits when each input SAR image was taken, the orbit that passes through the point closest to the center of all orbits is taken as the reference orbit. The difference between the reference orbit and the orbit when the SAR image of interest (nth SAR image) was taken is projected vertically in the line of sight from the satellite to the ground (vertical baseline length) and is called B. n Let α be a constant coefficient determined according to the angle of incidence, etc. In this case, the phase at pixel p of SAR image n is calculated as αB based on the height h at pixel p. n h p It is calculated as follows.

[0037] The difference in height between pixel p and its neighboring pixel (p+1) is Δh p =h p -h p+1 The parameter k that constrains the phase gradient between two adjacent pixels p is the appropriate constraint strength |k p | and the phase difference ∠k according to the height gradient p =αB n Δh p The strength of the constraint |k p For example, | may be set small for an analysis object that is likely to experience large thermal expansion or bend easily, and set large for an analysis object that hardly expands or bends.

[0038] [Method 2: Create a probability density function that constrains the phase difference between adjacent pixels (three or more pixels)] The above phase gradient model is a model created by focusing on the phase difference between two adjacent pixels. However, it is also possible to impose constraints on more adjacent pixels. For example, for pixel p and its adjacent m (m≧3) pixels p+1, p+2, . . . , p+m, the phase θ p ,θ p+1 ,θ p+2 ,···,θ p+m A complex vector s with → p =(e jθp ,e jθp+1 ,ejθp+2 ,···e jθp+m ) T The phase gradient is constrained by the evaluation function shown in equation (2) (called the multi-pixel phase gradient evaluation function) using the above. In equation (2), the arrow → indicates a vector. T indicates a transpose. H indicates a complex conjugate transpose.

[0039]

number

[0040] In equation (2), K p is a complex matrix. Complex matrix K p The elements of k in the above two-pixel case are p Similarly to (1), it is a complex number that includes the phase difference between pixels and a parameter that controls the strength of the constraint on the phase difference between pixels. In other words, the evaluation function in equation (2) corresponds to a function that expands the probability density distribution for two pixels to that for many pixels.

[0041] Compared to the phase gradient model, which constrains the phase difference between two pixels, the evaluation function in (2) can constrain higher-order differences, making it easier to constrain the phase gradient to have a smoother gradient and to constrain quantities such as curvature as the second derivative of the phase.

[0042] In addition, as in the case of restricting the phase difference between two pixels, when restricting the phase difference between multiple pixels, K p can be defined by the trajectory and the height at pixel p, etc. When the phase difference between multiple pixels is constrained, the curvature, etc. can also be constrained, so in addition to structures such as transmission towers, for example, in the case of a dome shaped like a split sphere, it becomes possible to evaluate the degree of match of the model according to the curvature of the dome ceiling and extract the phase of the dome ceiling.

[0043] The function is not limited to the form shown in equation (2). For example, Σ i a p,i θ p+i , or the unwrapped version W(Σ i a p,i θp+i ) may be constrained. p,i is a coefficient for calculating the phase gradient. p,-1 =-1, a p =2, a p,+1 If specified to be =-1, then Σ i a p,i θ p+i is the value equivalent to the second derivative of θ at p. i a p,i θ p+i By constraining the phase gradient, it can be constrained to a constant gradient. Also, since the phase returns to the same value when 2π is added, by applying a function W that converts the phase to a value that cycles between -π and π, such as by calculating the remainder when the phase is divided by 2π, W(Σ i a p,i θ p+i ) can be used to constrain the phase difference.

[0044] [Method 3: Create a template image] The above two methods create a matrix-based phase gradient model. However, a phase gradient model can also be created based on the concept of template matching. The template image used as the phase gradient model in Method 3 is an image related to the phase gradient that the object of analysis should exhibit.

[0045] For example, the input SAR image is in Let the template image be I template Then, the signal separation unit, which will be described later, may evaluate the analysis target using the following equation (3).

[0046]

number

[0047] In equation (3), "○" indicates the product of elements. Σ imagerepresents the sum over the entire image. g represents a monotonically increasing function. g outputs a large value when the phase gradients between the input SAR image and the template image match, and outputs a small value when the phase gradients differ. In other words, like Methods 1 and 2 above, Method 3 also essentially constrains the phase gradient.

[0048] I in and I template Alternatively, the local phase gradient may be evaluated by weighting the sum of neighboring pixels of a particular pixel.

[0049] The template image used may be an image taken in the past when no layovers or the like occurred, or an image taken in the past when a layover or the like occurred, from which the effects of the layover or the like have been removed. Therefore, when evaluating the object of analysis, trajectory information of the flying object or the three-dimensional shapes of structures other than the object of analysis are not required.

[0050] Embodiment 1. Fig. 2 is a block diagram showing an example of the configuration of an image analysis device according to the first embodiment. The image analysis device shown in Fig. 2 includes a division unit 110 that receives an SAR image from an SAR image storage unit 20, a phase gradient removal unit 120 that receives a phase gradient model related to a reflector to be analyzed from a phase gradient model storage unit 30, and a signal extraction unit 130. The phase gradient model storage unit 30 stores in advance a phase gradient created using one of the above-described techniques. The signal extraction unit 130 includes a matrix calculation unit 131 and a signal restoration unit 132.

[0051] The dividing unit 110 divides each SAR image into a plurality of small regions. Note that the small regions may partially overlap with other small regions. The phase gradient removing unit 120 removes the phase gradient of each small region from each SAR image in accordance with a phase gradient model.

[0052] The phase gradient removal unit 120 estimates the local phase gradient in each small region according to the phase gradient model, and removes the estimated phase gradient.

[0053] The matrix calculation unit 131 in the signal extraction unit 130 calculates a predetermined matrix C for each small region of each SAR image. Specifically, the matrix calculation unit 131 creates a matrix by adding together multiple pixels (complex numbers) from which the phase gradient has been removed. The signal restoration unit 132 extracts a dominant component (e.g., a principal component) from the created matrix. The dominant component corresponds to a signal indicating the phase of the analysis target. Therefore, the signal restoration unit 132 can obtain a signal related to the phase of the analysis target.

[0054] Next, the operation of the image analysis device of this embodiment will be described with reference to the flowchart of FIG.

[0055] The dividing unit 110 receives the SAR images stored in the SAR image storage unit 20. The dividing unit 110 divides each SAR image into small regions (step S101).

[0056] The phase gradient removal unit 120 removes the phase gradient of each small region for each SAR image according to the phase gradient model (step S102). Specifically, the phase gradient removal unit 120 removes the phase gradient that maximizes the evaluation value calculated based on the phase gradient model. The phase gradient removal unit 120 may remove the phase gradient that takes the expected value weighted by the evaluation value. The processing of step S102 removes, for example, a phase that depends on the height of the analysis target (transmission tower).

[0057] If the phase gradient model is created using the above method 3, i.e., if it is represented by a template image (template array), the phase gradient removal unit 120 shifts the phase of the entire template array in step S102 so that the phase at the location of the template array that coincides with the center of the small region becomes 0. The template array after the phase shift corresponds to the optimal value of the phase difference expected from the center of the small region with the surrounding pixels. The phase gradient is removed by multiplying each pixel by the complex conjugate of the template array after the phase shift.

[0058] When the phase gradient model is created by the above-mentioned method 1, that is, when it is expressed by a phase difference function between adjacent pixels, the phase gradient removal unit 120 performs the process of step S102 by using k p By integrating the phase differences of the pixels, the optimal value of the expected phase difference between the center of the small region and the surrounding pixels can be obtained. Thus, the phase gradient is removed, just as when the phase gradient model is represented by a template image.

[0059] The phase at the center of the small region is fixed to 0, and the probability density of the phases of the surrounding pixels can be calculated using the following (4).

[0060]

number

[0061] The phase gradient removal unit 120 calculates e jθp-d ~e jθp+d The phase gradient may be removed by calculating the expectation of , and multiplying pixel-wise by the complex conjugate of the expectation.

[0062] When the phase gradient model is created by the above-described method 2, i.e., when it is expressed by a multi-pixel phase gradient evaluation function, the phase gradient removal unit 120 removes the phase gradient using, for example, an optimal value in the process of step S102, in the same way as when the phase gradient model is created by the above-described method 1. The phase gradient removal unit 120 may remove the phase gradient from the multi-pixel phase gradient evaluation function by using, for example, an expected value of other pixels when the phase of the target pixel p is set to 0.

[0063] The matrix calculation unit 131 calculates a predetermined matrix C for each small region (step S103). The predetermined matrix C can be expressed as in the following equation (5). In equation (5), f is the average in the small region, the sum in the small region, x m ,x n The average of the small area after normalizing the absolute value of x to 1, or m ,x n It represents the sum in a small area after normalizing the absolute value of to 1.

[0064]

number

[0065] The signal restoration unit 132 calculates the dominant component (dominant phase change) in the matrix C (step S104). For example, in the process of step S104, the signal restoration unit 132 calculates y H The vector y that maximizes Cy and has a constant norm or the absolute value of all elements is 1 is calculated as the dominant component. The signal restoration unit 132 calculates y as the dominant component by using the element product of an appropriate real-number weighting matrix W and a matrix C according to the following equation (6): H You can also calculate a vector y that increases the value of Cy.

[0066]

number

[0067] The signal restoration unit 132 outputs the vector y as a complex vector having the phase of the analysis target (step S105). The vector y corresponds to a signal indicating the phase of the analysis target.

[0068] The image analysis device performs the processes of steps S101 to S105 for each of the N SAR images. Therefore, the image analysis device outputs N complex vectors (vector y). Using the N complex vectors, it is possible to perform displacement analysis using any displacement analysis method (for example, SBAS). The N complex vectors represent images formed from image signals in which the phase gradient of the analysis target has been removed from the SAR image according to a phase gradient model. In other words, the N complex vectors represent images in which the component of the analysis target has been emphasized. Therefore, the displacement analysis method can analyze the displacement of a reflector that exhibits weak reflection due to the influence of layover.

[0069] If a SAR image Note contains multiple analysis targets, the image analysis device performs the processes of steps S101 to S105 for N SAR images for each analysis target, thereby enabling analysis of the displacement of multiple reflectors affected by layover. Furthermore, while a linear phase change model (a time-series model of phase change) is prepared in advance in the time series analysis of the background art, a spatial phase gradient model is used in this embodiment. Therefore, this embodiment does not require the assumption of a linear model, and makes it possible to analyze nonlinear displacements of the analysis target.

[0070] Embodiment 2. Fig. 4 is a block diagram showing an example of the configuration of an image analysis device according to the second embodiment. The image analysis device shown in Fig. 4 includes a phase estimation unit 210 and an intensity estimation unit 220. The image analysis device according to the second embodiment uses a peripheral reflection model stored in a peripheral reflection model storage unit 40 and a reflection intensity a priori model stored in a reflection intensity a priori model storage unit 50, in addition to the SAR image stored in the SAR image storage unit 20 and the phase gradient model stored in the phase gradient model storage unit 30.

[0071] FIG. 5 is an explanatory diagram for explaining the reflection from an object to be analyzed (such as a power transmission tower) and the reflection from an area other than the object to be analyzed (referred to as a surrounding area) in the shooting area of the SAR image. σ indicates the reflection intensity of the object to be analyzed, and y → p is a complex vector containing the phase of the object to be analyzed at pixel p. ground corresponds to the ambient reflection model.

[0072] The peripheral reflection model is a model that represents the degree of reflection (intensity, noise pattern, etc.) from the peripheral area. The area other than the analysis target is an area that can be considered to be approximately flat, onto which the analysis target, such as a power transmission tower, could fall. The reflection intensity a priori model is a model that represents the degree of reflection (intensity, noise pattern, etc.) from the analysis target. In this embodiment, the image analysis device uses the peripheral reflection model and the reflection intensity a priori model to separate the phase of the analysis target from the phase of the peripheral area.

[0073] Let y be the complex vector with the phase of the object to be analyzed at pixel p. → p The complex reflection intensity of the object to be analyzed at pixel p is defined as σ p Let σ p represents the complex reflection intensity that should be exhibited by the actual observation target (transmission tower).

[0074] The observed signal (observed value) of pixel p is expressed by equation (7). N in equation (7) corresponds to the number of SAR images. The complex reflection intensity σ to be analyzed p Reflection intensity prior distribution model (reflection intensity prior model) p power (σ p ) is expressed by equation (8). In equation (8), σ without a subscript generally means the complex reflection intensity that is assumed to be exhibited by the object of observation. A complex vector with the phase of the object of analysis is expressed as y → p is expressed by equation (9). The prior information p of the phase gradient of the analysis object in the image with image number n grad,n (y n,1 ,y n,2 , ...) is expressed by equation (10). The prior information of the phase gradient corresponds to the phase gradient model. The peripheral reflection model p ground (x → g ) is expressed by equation (11).

[0075] The pixel values observed at pixel p are expressed as a vector for N images, and x → p Then, when the complex reflection intensity at pixel p and the complex vector with the phase of the object are observed superimposed, x → p , the distribution of layover areas p layover (x → p |σ p ,y → p ) is expressed by equation (12). In this embodiment, the layover area is an area in the SAR image where the analysis object and the ground surface around the analysis object overlap. → gis a complex vector corresponding to the reflection from the ground surface mixed with the reflection from the object of analysis.

[0076]

number

[0077] The optimization means (including the phase estimation unit 210 and the intensity estimation unit 220) calculates the complex vector y that maximizes the following equation (13): → p and complex reflection intensity σ p It is estimated that:

[0078]

number

[0079] Next, the operation of the image analysis device of this embodiment will be described with reference to the flowchart of FIG.

[0080] The phase estimation unit 210 estimates the peripheral reflection model p ground (x → p ) based on the distribution of layover areas p layover (x → p |σ p ,y → p ) and the phase gradient model p grad,n (y n,1 ,y n,2 ,…) and the pixel value x of the observed image → p complex vector y with optimal phase from → p (Step S201). Layover occurs when the influence of reflection from a power transmission tower and the influence of reflection from the ground surface overlap on a pixel in an SAR image. The intensity estimation unit 220 calculates the surrounding reflection model p ground (x → p ) based on the distribution of layover areas p layover (x → p |σ p ,y→ p ) and the reflection intensity prior model p power (σ p ) based on the optimal σ p is calculated (step S202).

[0081] The processes in steps S201 and S202 are performed to obtain the complex vector y → p and complex reflection intensity σ p The intensity estimation unit 220 repeats the process until it is determined that the complex vector y → p and complex reflection intensity σ p It is determined whether or not the values converge to the optimum values.

[0082] The intensity estimation unit 220 estimates the optimal complex vector y → p That is, the optimized complex vector y → p as a complex vector having the phase of the analysis target (step S203). p That is, the optimized complex reflection intensity σ p is output as the complex reflection intensity at the pixel p to be analyzed (step S203).

[0083] The phase estimation unit 210 calculates the complex vector y → p and complex reflection intensity σ p It may be determined whether or not the process of step S201 and the process of step S202 have converged to an optimal value. Also, the optimization means may be configured so that the process of step S201 and the process of step S202 are executed simultaneously.

[0084] In this embodiment, the optimization means calculates the complex vector y → p Optimal value of and complex reflection intensity σ p and convert them into a complex vector y → p Optimal value of and complex reflection intensity σ pHowever, the optimization method is to optimize the complex vector y → p and complex reflection intensity σ p The posterior distribution of is estimated, and the mean, variance, etc. of the posterior distribution are calculated as the optimized complex vector y → p and complex reflection intensity σ p However, since an analytical solution cannot be obtained when estimating the posterior distribution, the optimization means may be approximated as a set of samples using MCMC (Markov chain Monte Carlo methods) or the like. The optimization means may also be approximated as a function that can be handled analytically using the variational Bayes method or the like.

[0085] The optimization method is the Σ ground When performing calculations on the observed image pixel value x → p x is the direct product of itself and its complex conjugate → p x → p H Alternatively, the optimization means may calculate the average of Σ ground The optimization means may calculate Σ based on known information such as seasonal vegetation changes. ground may be estimated.

[0086] In this embodiment, the image analysis device uses a spatial phase gradient model, so that it is possible to analyze nonlinear displacements of the object to be analyzed, as in the first embodiment. Furthermore, the image analysis device uses a peripheral reflection model and a prior reflection intensity model to separate signals having a phase based on the reflection of the object to be analyzed from signals having a phase based on the reflection of the peripheral region. By using the peripheral reflection model, the performance of separating these two signals is improved. Furthermore, the obtained complex reflection intensity σ p This can be used to recognize the pattern of the object to be analyzed.

[0087] Embodiment 3. Fig. 7 is a block diagram showing an example of the configuration of an image analysis device according to the third embodiment. The image analysis device shown in Fig. 7 includes a layover area estimation unit 310 in addition to the components of the image analysis device according to the second embodiment shown in Fig. 4. The layover area estimation unit 310 estimates a layover area using a layover area continuity model stored in the layover area continuity storage unit 60.

[0088] If a pixel is in a layover area, there is a high possibility that neighboring pixels are also in the layover area. The layover area continuity model is a model that indicates that pixels neighboring pixels in a layover area are also in the layover area. When a layover area extends in one direction (the direction corresponding to the satellite line of sight) in an SAR image of a high-rise building or power transmission tower, a model of the continuity of the layover area in the satellite line of sight can be constructed as a Markov model, and layover areas can be detected using a forward backward algorithm or the like.

[0089] FIG. 8 is an explanatory diagram for explaining the layover area continuity model. FIG. 8(A) shows the same content as FIG. 5. For example, as shown in FIG. 8(B), a one-hot vector representing the class (whether it is a layover area or not) at pixel p is expressed as Z → p =(Z p,ground ,Z p,layover ) H Let's say Z → p is a variable that is 1 only for the corresponding class and 0 for the others. The final one-hot vector Z → p corresponds to the estimated layover area.

[0090] The layover area continuity model is, for example, p(Z → p |Z → p-1 )=Z → p T A-Z → p-1A is defined as follows. A is a matrix that represents the strength of continuity, and in row i and column j it stores the probability that pixel p is of class i when pixel (p-1) is of class j. In other words, as the diagonal elements of A increase, the probability that the class of pixel (p-1) and the class of pixel p are the same increases, strongly reflecting the assumption of continuity that adjacent pixels should be of the same class. In the layover area continuity model, the performance of detecting continuous layover areas is improved by setting a high probability that adjacent pixels p,(p-1) will have the same one-hot vector.

[0091] Furthermore, the area that should have a continuous phase is obtained as the layover area estimation result. As a result, the displacement analysis method using the output of the image analysis device of this embodiment shown in Figure 7 can more accurately perform the unwrapping process, which is a process of accumulating continuous phases, by also using the layover area estimation result.

[0092] When the observed values and various distributions of pixel p are defined in the same way as those in the second embodiment (see the above equations (7) to (12)), the optimization means (in this embodiment, including the phase estimation unit 210, the intensity estimation unit 220, and the layover region estimation unit 310) calculates a complex vector y having the phase of the analysis target at pixel p, which maximizes the following equation (14): → p and complex reflection intensity σ p and layover area Z → p Estimate.

[0093]

number

[0094] Next, the operation of the image analysis device of this embodiment will be described with reference to the flowchart of FIG.

[0095] The processing in steps S201 and S202 is the same as the processing in the second embodiment shown in FIG.

[0096] The layover area estimation unit 310 uses the layover area continuity model to calculate the optimal layover area Z → p is calculated (step S301).

[0097] The processes in steps S201, S202, and S301 are performed to obtain the complex vector y → p and complex reflection intensity σ p and layover area Z → p The process is repeated until it is determined that the complex vector y → p and complex reflection intensity σ p and layover area Z → p It is determined whether or not the values converge to the optimum values.

[0098] The layover area estimation unit 310 calculates the optimal complex vector y → p That is, the optimized complex vector y → p as a complex vector having the phase of the analysis target (step S302). p That is, the optimized complex reflection intensity σ p is output as the complex reflection intensity at the pixel p to be analyzed (step S302). → p That is, the optimized layover area Z → p is output (step S302).

[0099] Note that the phase estimation unit 210 or the intensity estimation unit 220 estimates the complex vector y → p and complex reflection intensity σ p and layover area Z → pIt may be determined whether or not the process of step S201, the process of step S202, and the process of step S301 have converged to an optimal value. Also, the optimization means may be configured so that the process of step S201, the process of step S202, and the process of step S301 are executed simultaneously.

[0100] In this embodiment, the optimization means calculates the layover area Z based on equation (14). → p However, the optimization means estimates the complex vector y → p Optimal value of and complex reflection intensity σ p As in the case of estimating the optimal value of the layover area Z → p The posterior distribution (expected value) of is estimated, and the mean and variance of the posterior distribution are used to estimate the optimal layover area Z → p It may be output as

[0101] Also, if the posterior distribution is used, the layover region Z → p is not expressed as a binary value of 1 or 0, but as a real number greater than 0 whose sum is 1, as shown in FIG. 8(C). In this case, Z → p Each component of represents the posterior probability of belonging to a respective class.

[0102] Embodiment 4. Fig. 10 is a block diagram showing an example of the configuration of an image analysis device according to the fourth embodiment. The image analysis device shown in Fig. 10 is obtained by adding a layover area estimation unit 410 to the image analysis device according to the first embodiment shown in Fig. 2.

[0103] Next, the operation of the image analysis device of this embodiment will be described with reference to the flowchart of FIG.

[0104] The processing in steps S101 to S104 is the same as the processing in the first embodiment shown in FIG.

[0105] In this embodiment, the signal extraction unit 130 or the layover area estimation unit 410 calculates the likelihood of a layover area based on the complex vector y having the phase of the analysis target obtained in the processing of step S104. Then, the layover area estimation unit 410 estimates the layover area using the layover area continuity model stored in the layover area continuity storage unit 60.

[0106] Specifically, the layover area estimation unit 410 calculates the y H Cy is regarded as the likelihood of a layover area, and the optimal layover area Z is calculated based on the layover area continuity model. p is calculated (step S401).

[0107] Then, the signal restoration unit 132 outputs the vector y as a complex vector having the phase of the analysis target (step S402). → p is output (step S402).

[0108] As in the third embodiment, in this embodiment, an area that should have a continuous phase is obtained as a layover area estimation result. As a result, the displacement analysis method using the output of the image analysis device of this embodiment shown in Figure 10 can more accurately perform the unwrapping process, which is a process of accumulating continuous phases, by also using the layover area estimation result.

[0109] As an example, the image analysis device according to each of the above embodiments can be used as follows: For example, by following the procedure below, it is possible to quickly check the condition of a power transmission tower when a flood or the like occurs.

[0110] First, the user specifies the general locations of power transmission towers across the country on a map. Then, the user extracts pixels from the SAR image around the power transmission tower and inputs them into the image analysis device. The image analysis device extracts the phase (i.e., displacement) of the power transmission tower using the method described above. A system including the image analysis device, or a system that has received the analysis results from the image analysis device, issues an alert (warning) when it detects that the ground beneath the power transmission tower has been eroded by a flood or other cause. The alert can then trigger a user to conduct a detailed inspection using other sensor data, etc.

[0111] Each component in each of the above embodiments can be configured as a single piece of hardware, or as a single piece of software. Each component can also be configured as multiple pieces of hardware, or as multiple pieces of software. Furthermore, some of the components can be configured as hardware, and the other parts can be configured as software.

[0112] Each function (each process) in the above-described embodiments can be realized by a computer having a processor such as a CPU (Central Processing Unit), a memory, etc. For example, a program for implementing the method (process) in the above-described embodiments may be stored in a storage device (storage medium), and each function may be realized by executing the program stored in the storage device by a CPU.

[0113] FIG. 12 is a block diagram showing an example of a computer having a CPU. The computer is implemented in an image processing device. The CPU 1000 executes processing in accordance with a program stored in a storage device 1001, thereby realizing the functions of the above-described embodiment. That is, it realizes the functions of the division unit 110, phase gradient removal unit 120, signal extraction unit 130, matrix calculation unit 131, signal restoration unit 132, phase estimation unit 210, intensity estimation unit 220, and layover region estimation units 310 and 410 in the image processing devices shown in FIGS. 2, 4, 7, and 10. Note that instead of the CPU 1000, a GPU (Graphics Processing Unit) or a combination of a CPU and a GPU can also be used.

[0114] The storage device 1001 is, for example, a non-transitory computer-readable medium. The non-transitory computer-readable medium includes various types of tangible storage media. Specific examples of non-transitory computer-readable media include magnetic recording media (e.g., hard disks), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Compact Disc-Read Only Memory), CD-Rs (Compact Disc-Recordable), CD-R / Ws (Compact Disc-Rewritable), and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), and flash ROMs). Furthermore, when a rewritable storage medium is used as the storage device 1001, the storage device 1001 can also be used as the SAR image storage unit 20, the phase gradient model storage unit 30, the surrounding reflection model storage unit 40, the reflection intensity prior model storage unit 50, and the layover region continuity storage unit 60.

[0115] The program may also be stored in various types of transitory computer-readable media, to which the program is supplied, for example, via a wired or wireless communication path, i.e., via an electrical signal, an optical signal, or an electromagnetic wave.

[0116] The memory 1002 is realized by, for example, a random access memory (RAM), and is a storage means for temporarily storing data when the CPU 1000 executes processing. A configuration is also conceivable in which a program held in the storage device 1001 or a temporary computer-readable medium is transferred to the memory 1002, and the CPU 1000 executes processing based on the program in the memory 1002.

[0117] Fig. 13 is a block diagram showing the main components of an image analysis device. The image analysis device 1 shown in Fig. 13 includes a signal separation means 2 (implemented by a division unit 110, a phase gradient removal unit 120, and a signal extraction unit 130, or a phase estimation unit 210 and an intensity estimation unit 220 in the embodiment) that receives as input a plurality of radar images capturing the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and extracts from the radar images components that match the phase difference represented by the phase gradient model.

[0118] The signal separation means 2 may include a division means (realized by the division unit 110 in the embodiment) that divides the radar image into small regions, a phase gradient removal means (realized by the phase gradient removal unit 120 in the embodiment) that removes components that match the phase difference represented by the phase gradient model from each small region, and a signal extraction means (realized by the signal extraction unit 130 in the embodiment) that extracts dominant components in the radar image from which the phase difference has been removed.

[0119] The signal separation means 2 may include a phase estimation means (realized by the phase estimation unit 210 in the embodiment) that estimates the phase difference represented by the phase gradient model in the radar image using a peripheral reflection model and a phase gradient model that represent the distribution of reflection from a peripheral area other than the area to be analyzed, and an intensity estimation means (realized by the intensity estimation unit 220 in the embodiment) that estimates the reflection intensity from the area to be analyzed using a reflection intensity prior model and a phase gradient model that represent the distribution of reflection from the area to be analyzed.

[0120] The image analysis device 1 may also be equipped with a layover area estimation means (in an embodiment, realized by layover area estimation units 310, 410) that estimates the area to be analyzed using a layover area continuity model that indicates that pixels adjacent to the pixel to be analyzed are also likely to be the pixel to be analyzed.

[0121] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention. [Explanation of symbols]

[0122] 1. Image analysis device 2 Signal separation means 10 Signal separation section 20 SAR image storage unit 30 Phase gradient model storage section 40 Surrounding reflection model storage section 50 Reflection intensity prior model storage section 60 Layover Area Continuity Storage 110 Split section 120 Phase gradient removal unit 130 Signal Extraction Unit 131 Matrix calculation section 132 Signal restoration unit 210 Phase estimation section 220 Intensity estimation part 310,410 Layover Area Estimation Unit 1000 CPU 1001 Storage device 1002 memory

Claims

1. a signal separation means for receiving a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and extracting from the radar images components that match the phase difference represented by the phase gradient model; The signal separating means a dividing means for dividing each of the plurality of radar images into small regions; a phase gradient removing means for removing, from each small region, a component that matches the phase difference represented by the phase gradient model; and a signal extraction means for extracting a dominant component in the radar image from which the phase difference has been removed. Image analysis device.

2. An image analysis device that outputs a phase difference and a reflection intensity, a signal separation means for receiving a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and extracting from the radar images components that match the phase difference represented by the phase gradient model; The signal separating means a phase estimation means for estimating a phase difference represented by the phase gradient model in a radar image using a peripheral reflection model representing a distribution of reflections from a peripheral region other than the region to be analyzed and the phase gradient model; an intensity estimation means for estimating a reflection intensity from the analysis object using a reflection intensity prior model representing a distribution of reflection from the analysis object and the phase gradient model; An image analysis device comprising:

3. A layover area estimation means is provided for estimating the area of the analysis target using a layover area continuity model that indicates that pixels adjacent to the pixel of the analysis target are also highly likely to be the pixel of the analysis target.

3. The image analysis device according to claim 1.

4. inputting a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and extracting components from the radar images that match the phase difference represented by the phase gradient model; Dividing each of the plurality of radar images into small regions; removing from each subregion a component that matches the phase difference represented by the phase gradient model; Extracting the dominant component in radar images with phase differences removed Image analysis methods.

5. An image analysis method that outputs a phase difference and a reflection intensity, inputting a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and extracting components from the radar images that match the phase difference represented by the phase gradient model; using a peripheral reflection model that represents a distribution of reflections from a peripheral region other than the region to be analyzed and the phase gradient model, to estimate a phase difference represented by the phase gradient model in the radar image; The reflection intensity from the analysis object is estimated using the phase gradient model and a reflection intensity prior model that represents a distribution of reflection from the analysis object. Image analysis methods.

6. The area to be analyzed is estimated using a layover area continuity model that indicates that pixels adjacent to the pixel to be analyzed are also likely to be the pixel to be analyzed. The image analysis method according to claim 4 or 5.

7. On the computer, a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and a process of extracting components that match the phase difference represented by the phase gradient model from the radar images; Dividing each of the plurality of radar images into small regions; removing from each subregion a component that matches the phase difference represented by the phase gradient model; Extracting the dominant components in a radar image with phase differences removed Image analysis program for.

8. A computer comprising: a plurality of radar images of the same area and a phase gradient model representing the phase difference between a plurality of adjacent pixels on the surface of an object to be analyzed that may be present in the radar images, and a process of extracting components that match the phase difference represented by the phase gradient model from the radar images; using a peripheral reflection model that represents a distribution of reflections from a peripheral region other than the region to be analyzed and the phase gradient model, to estimate a phase difference represented by the phase gradient model in the radar image; estimating a reflection intensity from the analysis object using a reflection intensity prior model representing a distribution of reflection from the analysis object and the phase gradient model; The estimated phase difference and the estimated reflection intensity are output. Image analysis program for.

Images