JP2010175381A - Image change extraction apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image change extraction apparatus for extracting a physical change occurring on the surface of the earth with a higher accuracy and for correctly displaying the spacial continuity of the change. <P>SOLUTION: The image change extraction apparatus includes, a noise removal unit 2 for removing the noise targeting a coherence map, a characteristic amount extraction unit 6 for extracting the characteristic amount from the image after the noise removal, a coherence average calculation unit 3 for calculating the temporal and spacial average of pixels composing the coherence map, a change extraction unit 4 for extracting the change using time-series change information of the coherence map, a change time adjustment unit 5 for detecting the discrepancy of the occurrence time of the change and for making the time adjustment, and a change sorting unit 7 for sorting the change and outputting the change sorted image data after the change sorting as an overall output. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an image change extraction apparatus that extracts physical changes from a plurality of complex correlation distributions (coherence maps) obtained from a plurality of radar images obtained by a plurality of periodic observations.

  It occurred on the ground surface between the time when the first image was acquired and the time when the second image was acquired from two radar images with different acquisition times for the same area obtained under almost the same conditions. One technique for extracting changes is CCD (Coherent Change Detection). In CCD, since change detection is performed using phase information in addition to luminance information of a SAR (Synthetic Aperture Radar) image, there is a feature that sensitivity to the change is higher than a change extraction technique based on luminance change.

  In a CCD, a complex correlation value (coherence) between two SAR images is used as an index representing a change occurring on the ground surface. This coherence has a property of taking a value close to 1 in a region where there is no change and taking a value close to 0 in a region where there has been a change. Therefore, as for the change in the ground surface, it is possible to extract a change area by performing threshold processing on an area with low coherence between two SAR images, that is, extracting an area below the threshold. For example, in the prior art, threshold processing based on a false alarm probability set from a coherence distribution (coherence map) calculated from two SAR images is performed (for example, see Non-Patent Documents 1 to 3).

Mark Preiss, Doug Gray, and Nick Stacy, "A Change Detection Technique for Repeat Pass Interferometric SAR," IEEE Geoscience and Remote Sensing Symposium Proceedings, 2003. IGARSS '03. 2003, 2, 21-25 Jong-Sen Lee, Karl W. Hoppel, Stephen A. Mango and Allen R. Miller, "Intensity and Phase Statistics of Multilook Polarimetric and Interferometric SAR Imagery", IEEE Transaction On Geoscience and Remote Sensing, vol.32, No.5, 1994 Howard. A. Zebker, J. Villasenor, "Decorrelation in interferometric Radar Echoes," IEEE Transaction On Geoscience And Remote Sensing, Vol.30, pp.950-959, 1992.

  In the conventional image change extraction apparatus, a change region is extracted by threshold processing from a coherence map between two SAR images. However, the extracted changes include changes that are unnecessary for the operator (interpretation engineer), such as fluctuations in plants, and it is necessary for the operator to manually select the changes to be analyzed. Further, it is known that the coherence between two SAR images is constantly low in a region where minute changes continuously occur, such as a water area and a grassland. For this reason, when a change occurs repeatedly in the steady change region, even if there is a difference in the coherence distribution itself, it is difficult to extract the change in the process using the lower limit value of the threshold in the conventional image change extraction device. Further, even if an overlap change is extracted by an operator or the like, there is a problem that the change extraction timing is shifted and the spatial continuity of the change cannot be extracted correctly.

  The present invention has been made to solve the above problems, and can extract physical changes occurring on the ground surface with higher accuracy and correctly display the spatial continuity of the changes. An object is to provide a possible image change extraction device.

  The image change extraction apparatus according to the present invention stores N-1 coherence maps, which are distributions of coherence between two images, based on images obtained by observing N different times for the same region. Noise that is a coherence map file and a coherence map that is a complex correlation distribution between two images stored in the coherence map file is input and noise is removed from the coherence map. A noise removal unit that outputs removed image data, and a coherence map group that is a collection of coherence maps between two images stored in the coherence map file, and a temporal and space of pixels constituting the coherence map Average value is calculated and coherence map is used as coherence average data A coherence average calculation unit for outputting to a file, a coherence map between two images stored in the coherence map file, and coherence average data calculated from a plurality of coherence maps, and coherence average data; Using the time-series change information of the coherence map, change extraction is performed, change occurrence time data indicating the change occurrence time of the change occurrence area, change continuation status data that is information on whether or not the change occurrence is constantly occurring, and A change extraction unit that outputs change extraction data composed of upper and lower limit information data that is information of a determination value used for change extraction and change extraction pixel data that represents a change extraction region; and a change output from the change extraction unit Using the extracted data as input, change from the change occurrence time data, change continuation status data, and upper / lower limit information data. Is a change time adjustment unit that detects a shift in the occurrence time of the error and outputs post-time adjustment change extraction data that is change extraction data after the time adjustment, and a parameter relating to a change method to be selected and a selection target change region Change definition parameter file storing change definition parameters, noise-removed image data output from the noise removal unit, post-time adjustment change extraction data output from the change time adjustment unit, and change definition parameter file The change definition parameter stored in is input, the feature quantity extraction method is determined based on the change definition parameter, the feature quantity is extracted, and the feature quantity extraction result and the time-adjusted change extraction data are extracted. A feature amount extraction unit that outputs change definition parameters, feature amount extraction data output from the feature amount extraction unit, and time Input the adjusted change extraction data and change definition parameters, determine the change selection method based on the feature extraction data and change definition parameters, select the changes from the feature extraction data and the time-adjusted change extraction data, Change selection comprising change occurrence time data indicating the change occurrence time, change continuation status data indicating the chronological continuation status of the change, and change selection pixel data consisting of a change selection area and an ID uniquely representing the change selection area A change selection unit that stores data as a whole output in a change selection data file.

  According to the present invention, a physical change is extracted with higher accuracy from a plurality of complex correlation distributions (coherence maps) obtained from a plurality of radar images obtained by a plurality of periodic observations, and the spatial continuity of the changes is extracted. It becomes possible to display the sex correctly.

It is a block block diagram which shows the image change extraction apparatus by Embodiment 1 of this invention. It is a figure of the example of a change condition of the unsteady change area | region by Embodiment 1 of this invention. It is a figure of the example of a change of the coherence of the pixel of interest located in the area | region where the transient change on the non-stationary change area | region occurred by Embodiment 1 of this invention. It is a figure of the example of a change condition of the steady change area | region by Embodiment 1 of this invention. It is a figure of the example of a change of the coherence of the focused pixel located in the area | region where the transient change on the steady change area | region generate | occur | produced according to Embodiment 1 of this invention. It is a figure of the example of a change detection result before the time adjustment by Embodiment 1 of this invention. It is a figure of the example of a change detection result after time adjustment by Embodiment 1 of this invention. It is a block block diagram which shows the image change extraction apparatus by Embodiment 2 of this invention. It is a block block diagram which shows the image change extraction apparatus by Embodiment 3 of this invention. It is a block block diagram which shows the image change extraction apparatus by Embodiment 4 of this invention. It is a block block diagram which shows the image change extraction apparatus by Embodiment 5 of this invention. It is a block block diagram which shows the image change extraction apparatus by Embodiment 6 of this invention.

Embodiment 1 FIG. (Basic configuration)
1 is a block diagram showing an image change extraction apparatus according to Embodiment 1 of the present invention. The image change extraction apparatus shown in FIG. 1 stores N-1 coherence maps, which are distributions of coherence between two images, based on images obtained by observing N different times for the same region. A coherence map file 1 and a coherence map that is a complex correlation distribution between two images stored in the coherence map file 1 are input, noise is removed from the coherence map, and the image is after noise removal. The noise removal unit 2 that outputs noise-removed image data and a coherence map group that is a collection of coherence maps between two images stored in the coherence map file 1 are used as inputs, and the temporal relationship of pixels constituting the coherence map The spatial average is calculated and the coherence map data is used as the coherence average data. The coherence average calculation unit 3 that outputs to the image, the coherence map between the two images stored in the coherence map file 1, and the coherence average data calculated from a plurality of coherence maps are input. Using the time series change information of the coherence map, change extraction is performed, change occurrence time data indicating the change occurrence time of the change occurrence area, and change continuation status data indicating whether the change occurrence is constantly occurring And a change extraction unit 4 that outputs change extraction data composed of upper and lower limit information data, which is information of determination values used for change extraction, and change extraction pixel data representing a change extraction region.

  In addition, the change extraction data output from the change extraction unit 4 is input, the change occurrence data is detected from the change occurrence time data, the change continuation status data, and the upper / lower limit information data, and the change extraction data after the time adjustment is detected. A change time adjustment unit 5 that outputs time-adjusted change extraction data, a change definition parameter file 8 that stores a change definition parameter that is a parameter related to a selection target change method and a selection target change region, The noise-removed image data output from the noise removing unit 2, the time-adjusted change extraction data output from the change time adjusting unit 5, and the change definition parameter stored in the change definition parameter file 8 are input. The feature quantity extraction method is determined based on the definition parameter, the feature quantity is extracted, the feature quantity extraction data as the feature quantity extraction result, and the time The feature amount extraction unit 6 that outputs the modified change extraction data and the change definition parameter, the feature amount extraction data output from the feature amount extraction unit 6, the time-adjusted change extraction data, and the change definition parameter are input. Based on the feature extraction data and the change definition parameter, the change selection method is determined, the change is selected from the feature extraction data and the time-adjusted change extraction data, the change occurrence time data indicating the change occurrence time, The change selection data including the change continuation status data indicating the chronological continuation status and the change selection pixel data including the change selection area and the ID uniquely indicating the change selection area is stored in the change selection data file 9 as an overall output. And a change sorting unit 7 for performing the above.

  Next, the operation will be described. In the coherence map data file 1, the two images for calculating the coherence do not necessarily have adjacent observation times, but in the following, a coherence map calculated between two radar images having the adjacent observation times is used. An example in the case of storing will be described.

  The noise removal unit 2 receives a coherence map that is a complex correlation distribution between two images stored in the coherence map file 1 and outputs noise-removed image data that is an image after noise removal. The noise removing unit 2 removes pixel information unnecessary for change extraction / selection as noise from the coherence map. Here, an example in which moving average processing is performed as a noise removal method is shown. In the moving average process, the value of a certain pixel is set to the average value of surrounding values. Replacing with an average value has the effect of blurring noise, but on the other hand, the outline is also blurred.

  Instead of the moving average process, a noise removal process using a median filter may be performed. In the median filter, the values around a certain pixel are sorted and set to the median value. There is an advantage that only the noise is blurred without blurring the outline.

  In addition to the averaging process, the noise removal algorithm includes a contraction process for thinning the figure by one pixel and an expansion process for expanding the figure to the outside by one pixel. By the expansion / contraction process, it is possible to remove a hole corresponding to one pixel formed in a figure due to the influence of noise and a protrusion such as a whiskers.

  Next, the coherence average calculation unit 3 receives a coherence map group that is a collection of coherence maps between two images stored in the coherence map file 1 as an input, and averages the coherence of all the pixels constituting the coherence map. The value is output as coherence map average data and stored in the coherence map file 1. The coherence average calculation unit 3 calculates the time and space average values, which are the average values in the time and space directions, for all the pixels constituting the coherence map, and sets them as coherence average data.

  First, with respect to all the coherence maps stored in the coherence map storage unit, the average value M of the coherence of the target pixel on the map is calculated, and the time average which is the average value in the time direction is obtained. Further, by calculating an average value for a plurality of pixels including the target pixel and its neighboring pixels, the time and space average values are obtained. It is good also as only a time average, without being limited to the combination of a time average and a space average. As examples of the neighborhood, 4 neighborhoods and 8 neighborhoods are conceivable.

  Next, the change extraction unit 4 receives, as input, a coherence map between two images stored in the coherence map file 1 and coherence average data, , Change continuation status data that is information as to whether or not a change has occurred steadily, upper and lower limit information data indicating whether the threshold used for change extraction is an upper or lower limit, and a change extraction area Change extraction data consisting of change extraction pixel data is output.

  First, the change extraction unit 4 performs the steady change region determination based on the coherence average data. Specifically, the time and space average M, which is the average value of the time and space coherence of the pixel of interest, and the steady change pixel determination value T, which is a preset value, are compared. When it is low, it is defined as a steady change pixel, and the steady change determination result as to whether it is a steady change pixel is represented by a binary value of 1 or 0, for example, and the combination of the steady change determination result and the pixel ID of the pixel of interest continues to change Use situation data. In addition, a connected component composed of steady change pixels is defined as a steady change region.

  Next, threshold processing is performed on the coherence map, whether each pixel is changed or not is determined, a pixel determined to be changed is set as a change extraction region constituting pixel, and changes are made from a connected component of the change extraction region constituting pixel. An extraction area is defined, an ID uniquely determined for each change extraction area is assigned as a labeling value, and a pair of change extraction area constituting pixel and labeling value is set as change extraction pixel data.

  As a first example of threshold processing, a case where a change occurs in a region that is not a steady change region, that is, an unsteady change region is shown. When the coherence is equal to or less than a preset threshold value S, it is determined that a change has occurred. FIG. 2 shows an example of a change state on the non-stationary change region, and FIG. 3 shows an example of a change in the coherence value of the pixel of interest located in the region where the change has occurred on the non-stationary change region. In FIG. 2, a square area represents a non-steady change area, a black circle represents a pixel of interest, a shaded area represents a change occurrence area, and T1 to Tn + 1 represent time.

As an example of the threshold value setting method, the threshold value is determined by Expression (1) from the average value M of the coherence of the target pixel which is a non-stationary change pixel and ΔS which is a difference value specified in advance.
S = M−ΔS (1)

  The setting of the threshold value may be determined by a theoretical value of variance from the average value M of the coherence of the target pixel, not by the method using the difference value.

  As a second example of threshold processing, a transient change is made on the steady change region, “changes that do not change for a while after the change has occurred, or slight changes that do not require extraction”. An example of extracting transient changes is shown. A specific example is the extraction of traces on a meadow.

  As a first step, the coherence threshold value (upper limit Su and lower limit Sd) of the steady change pixel is determined from the average value M of the coherence of the pixel of interest which is the steady change pixel.

As an example of the threshold value setting method, the threshold value is determined as shown in Expression (2) from the difference value ΔS specified in advance, as in the case of change extraction on the unsteady change region.
Sd = M−ΔS, Su = M + ΔS (2)

  Alternatively, similarly to the case of change extraction on the non-stationary change region, it may be determined by the theoretical value of variance from the average value M of the coherence of the target pixel.

  Next, in the second step, it is determined that a change has occurred when the coherence of the pixel of interest constituting the transient change region becomes equal to or lower than the lower limit value Sd or equal to or higher than the upper limit value Su due to the occurrence of the temporary change. FIG. 4 shows an example of a change situation on the steady change region, and FIG. 5 shows an example of a change in the coherence value of the pixel of interest located in the region where the transient change has occurred on the steady change region. In FIG. 4, a square area represents a steady change area, a black circle represents a pixel of interest, a shaded area represents a transient change area, and T1 to Tn + 1 represent time.

  However, if the situation of Su or higher continues, it is regarded as a transient non-stationary change area that is a temporary non-steady change area, and only the first time while Su or higher is continuous. Is determined to have occurred, and thereafter no change is assumed. When the coherence decreases again, the transient unsteady change region is canceled and the steady change region is re-determined.

  In addition, in the upper / lower limit information flag, which is a flag indicating whether the value is lower limit value or less or upper limit value or more, for example, 1 is set if the value is not less than the upper limit value, and -1 is set if the value is not more than the lower limit value. Together with the lower limit information flag and the pixel ID of the target pixel, the upper / lower limit information data is obtained.

  As the change occurrence time, the timing at which the coherence of the pixel of interest falls outside the range of the threshold is set as the change occurrence time initial value in the change occurrence time data.

  In addition, in the second example of threshold processing, a change extraction example in the case where there is an overlapping change, i.e., a change in the steady change region, is shown. The change of the region itself can be extracted by the extraction method using two threshold values of the upper limit / lower limit shown in the second example of threshold processing. The change in the steady change region itself means the elimination of the factor that becomes the steady change region, specifically, the cutting of forest in the vegetation region, and the like.

  Next, the change time adjustment unit 5 receives the change extraction data output from the change extraction unit, and outputs post-time adjustment change extraction data that is change extraction data after time adjustment. The change time adjustment unit 5 detects a deviation of the change occurrence time from the change occurrence time data, change continuation status data, and upper / lower limit information data input as change extraction data, and performs time adjustment. Specifically, a steady change pixel is selected from the change continuation status data, and further, a case where the threshold upper limit is used for determination is selected from the upper and lower limit information data, and is set as a time adjustment target pixel. Specifically, the time one time stamp before the change occurrence time given as the change occurrence time data is used as the adjusted change occurrence time data that is the adjusted change occurrence time, and the change extraction pixel data and the change continuation The situation data and the change occurrence time data after the time adjustment are combined to obtain the change extraction data after the time adjustment.

  FIG. 6 shows an example of a change extraction result before time adjustment, and FIG. 7 shows an example of a change extraction result after time adjustment. Of the two consecutive rectangular regions, the left side is an unsteady change region, the right side is a steady change region, the shaded region is a transient change region, and T2 and T3 indicate time. The unsteady change region is assumed to be soil or mud, the steady change region is assumed to be grassland, and the trace is assumed to be a temporary change. The two small square areas in FIG. 7 are the same as the square areas in FIGS. In FIG. 6 before the time adjustment, changes that should be detected at the same time are detected at different times T2 and T3, and the result is as if different changes occurred at the two times. On the other hand, in FIG. 7 after the time adjustment, changes in the non-steady change region and the steady change region are represented in a continuous form, and the spatial continuity of the change is correctly displayed.

  For pixels that are not subject to time adjustment, change occurrence time data is used as change occurrence time data after time adjustment as it is, and change extraction pixel data, change continuation status data, The change occurrence time data after time adjustment is combined with the change extraction data after time adjustment.

  Next, the feature amount extraction unit 6 stores the noise-removed image data output from the noise removal unit 2, the post-time adjustment change extraction data output from the change time adjustment unit 5, and the change definition parameter file 8. The change definition parameter, which is a parameter related to the selection target change area, and the change definition parameter that is a parameter related to the selection target change, are input as feature value extraction data, time-adjusted change extraction data, and change definition. Parameter.

  The change definition parameters include “change pattern (rotation, vibration, repetition of the same pattern, irregular)” and other parameters related to the change method, “area, length, width”, “outline / shape (straight line, curve, circle) , Rectangle, irregular, etc.) ”.

  The feature quantity extraction unit 6 selects a feature quantity extraction method based on the change definition parameter and performs feature quantity extraction. A feature amount extraction method plan is set in advance for each change definition parameter. In this embodiment, first, the area calculation and aspect ratio calculation of a change area candidate are performed from among a plurality of feature amount extraction method plans. The case where it selects as an extraction method is shown.

  Based on the change extraction data output from the change extraction unit 4, the area and aspect ratio are calculated from the number of change extraction pixels constituting the change extraction region. The size per pixel is estimated based on imaging information such as resolution.

  Note that the feature quantity extraction method can be selected from a feature quantity extraction method proposal according to a recommended order set in advance for each change definition parameter, or by presenting a method proposal to the user, You may choose.

  Next, an example in which thinning processing is performed instead of using area and aspect ratio calculation as a feature amount extraction method will be described. The thinning process is a process of converting to a line image having a width of 1 pixel while maintaining the continuity of the figure without changing the topology of the figure, i.e., without breaking the line in the middle or without a hole. . The center line of the figure is extracted as a feature by thinning. This is considered effective when the shape of the center line of the change region is characteristic.

  Instead of the thinning process, a contour extraction process that is a process of extracting the contour of the change candidate region may be performed. As an example of the contour extraction method, there is a Laplacian filter which is a secondary differential filter.

  Next, the change selection unit 7 receives the feature amount extraction data output from the feature amount extraction unit 6, the change extraction data after time adjustment, and the change definition parameter, and outputs change selection data. Store in file 9. The change selection data includes time adjustment change occurrence time data, change continuation status data, and change selection pixel data. Furthermore, the change selection pixel data includes a change selection region constituent pixel that is the number of a pixel that forms the change selection region, and a labeling value that is an ID that uniquely represents the change selection region.

  The change definition parameters include “change pattern (rotation, vibration, repetition of the same pattern, irregular)” and other parameters related to the change method, “area, length, width”, “outline / shape (straight line, curve, circle) , Rectangle, irregular, etc.) ”.

  The change selection unit 7 selects a change selection method based on the change definition parameter and the feature amount extracted by the feature amount extraction unit 6, and selects changes from the feature amount extraction data and the time-adjusted change extraction data. A change selection method proposal is set in advance for each change definition parameter, and in the present embodiment, first, a threshold processing for the area and aspect ratio of the change area candidate is selected from a plurality of change selection method proposals. The case where it selects as a change selection method is shown.

  The area A and the aspect ratio R of the change extraction region output as the feature amount extraction data from the feature amount extraction unit 6 are compared with preset threshold values Sf1 and Sf2. For example, when the conditions of Sf1 <A and Sf2 <R are satisfied, the change extraction area is selected as a change selection area.

  The change selection region constituting pixels which are the numbers of the pixels constituting the change selection region and the corresponding labeling values are determined as change selection pixel data.

  After the change selection, the change occurrence time data after the time adjustment, the change continuation status data, and the change selection pixel data that is a combination of the change selection region pixel number and the labeling value are combined and output as change selection data.

  As the change selection method, a linear discriminant method that is a method for determining a boundary line may be applied. Examples of the linear discriminant include a linear discriminant based on a covariance matrix and a support vector machine (SVM). The linear discriminant method based on the covariance matrix projects data onto a vector w that passes through the cluster center representing the center of all data so that different types of data are separated as much as possible and the same type of data is distributed as close as possible. This is a method for discriminant analysis. In addition, the support vector machine is a method of obtaining a straight line that is a boundary based on the distance and direction from all the teacher data, and the distance is maximized and the direction of the straight line seen from the teacher data is the same for each type. This is a method for determining a straight line. When the number of attributes is N-dimensional, the boundary is a hyperplane. In the support vector machine, a nonlinear boundary can be defined by expressing a function necessary for learning by a linear sum of functions called a kernel function.

  Consider an example using a support vector machine. The support vector machine learns changes to be selected and sample data of non-selection target changes, and determines a selection target change and a non-selection target change by hyperplane calculation for discrimination. In addition to calculating the hyperplane based on the selection target change and the non-selection target change, when the extraction change is divided into a plurality of clusters, the multi-class classification problem may be determined by a tournament method or the like.

  In addition to the linear discriminant method, selection by a decision tree, a neural network, a Bayes classifier, or a K nearest neighbor method may be performed. The decision tree is a learning method that divides teacher data into subsets based on the value of each attribute, takes a tree structure, and represents a collection of features from leaves to classifications and branches to the classifications. A neural network is a learning method that imitates the neural network of life, and finally determines the optimum boundary between types based on teacher data. The Bayes classifier is a classification model that assumes that the feature vectors of each class have a normal distribution. The K nearest neighbor method is a method of determining a class from K neighborhood points of learning data.

  As described above, according to the first embodiment, change extraction is performed using time series information of change, and the change is steadily performed by correcting the spatial continuity of the change caused by the deviation of the extraction timing. It is possible to extract a change that occurs in an overlapping manner on the region where the occurrence of the error occurs and to correctly display the spatial continuity of the change.

  Further, by adding a change selection function, it is possible to automatically select changes to be analyzed, or to support change selection by the user.

Embodiment 2. FIG. (Use of ground truth information)
In the first embodiment, the noise removal unit 2 has shown the case of removing noise by the noise removal algorithm. However, when the terrain information of the target area is known to some extent in advance, the ground truth information that is the terrain information of the imaging area is shown. The case where pre-processing of change selection is performed using is shown.

  FIG. 8 is a block diagram showing an image change extraction apparatus according to Embodiment 2 of the present invention. In FIG. 8, the same parts as those shown in FIG. In the second embodiment shown in FIG. 8, a ground truth information file 10 storing ground truth information is further provided to the configuration of the first embodiment, and the change selection unit 7 outputs the feature value extraction unit 6. Feature extraction data, time-adjusted change extraction data, change definition parameters, and ground truth information stored in the ground truth information file 10 are input. Position information of the change, information such as shape / area, etc., and collating these information with information such as the position information of the time-adjusted change extraction pixel data and the area of the feature amount extraction data, and the comparison result is If they match, it is removed as a change outside the selection target.

  As a collation method, for example, a deviation width and an area difference between vertical and horizontal positions based on position information are calculated.

  It is assumed that the registration process that is the alignment between the coherence map data and the image data for which ground truth information is obtained has already been performed.

  As described above, according to the second embodiment, as a pre-process for change selection, an efficient change selection process based on actual terrain information is performed by removing non-selection changes based on ground truth information. Is possible.

Embodiment 3 FIG. (Change definition parameter classification storage)
In the first embodiment, the change definition parameter is used as it is, and change extraction, feature amount extraction, and change selection are exemplified. However, in this embodiment, the change definition parameter is divided into change classification classes and stored in the database. An example is shown.

  FIG. 9 is a block diagram showing an image change extraction apparatus according to Embodiment 3 of the present invention. In FIG. 9, the same parts as those shown in FIG. In the third embodiment shown in FIG. 9, in the configuration of the first embodiment, a change classification class file 11 that stores parameters representing a change classification class that is a classification of a change selection object, and the change classification class file 11 The stored change classification class parameter is used as the first input, the change definition parameter stored in the change definition parameter file 8 is used as the second input, and the change definition parameter is output separately for each change classification class. And a change definition parameter classification storage unit 13 that stores the change definition parameter data for each change classification class as change definition sample data in the change definition parameter classification database 12.

  The change classification class parameter is set by the user together with the change definition parameter, and may be a general name such as “river” or “轍”.

  As described above, according to the third embodiment, the change definition parameters are divided into change classification classes and stored in the database, so that correspondence cases between the change definition parameters and the change classification classes are accumulated, and the correspondence analysis is performed. It becomes possible to use it.

Embodiment 4 FIG. (Change classification class update)
In the third embodiment, an example in which the change definition parameter is divided for each change classification class set in advance is shown. However, in this embodiment, an example in which a change classification class is determined by learning is shown.

  FIG. 10 is a block diagram showing an image change extraction apparatus according to Embodiment 4 of the present invention. 10, parts that are the same as the parts shown in FIG. 1 are given the same reference numerals, and explanation thereof is omitted. In the fourth embodiment shown in FIG. 10, in contrast to the configuration of the first embodiment, the change definition parameter classification database 12 that stores change definition sample data for each change classification class and the changes stored in the change definition parameter file 8 are used. The definition parameter is the first input, the change definition sample data for each change classification class stored in the change definition parameter classification database 12 is the second input, the change classification class is learned from the change definition sample parameter, and a new A change classification class updating unit 14 that further determines a change classification class, updates the class classification of the change definition parameter classification database 12, and outputs the change classification parameter classification database 12 to the change definition parameter classification database 12.

  The change classification class update unit 14 receives the change definition sample data stored in the change definition parameter classification database 12 and the change definition parameter stored in the change definition parameter file 8 as input, and uses the change definition sample data as learning data. Reclassify the change classification class. The change definition parameters are divided for each new change classification class, the change definition sample data is updated, and is output to the change definition parameter classification database 12.

  As a classification method, an expected value maximization method known as a general classification method or a K-means method may be applied.

  As described above, according to the fourth embodiment, by learning and updating the change classification class based on the change definition parameter, the change classification according to the analysis content is not limited to the classification of the change name or the like. And change extraction based on it becomes possible.

Embodiment 5 FIG. (Automatic determination of change definition parameters)
In the first embodiment, the case where the user designates the change definition parameter in advance has been shown. However, in this embodiment, the database information set in advance is used from the change classification class name representing the classification of the selection target. The change definition parameter is determined.

  FIG. 11 is a block diagram showing an image change extraction apparatus according to Embodiment 5 of the present invention. In FIG. 11, the same parts as those shown in FIG. In the fifth embodiment shown in FIG. 11, a change classification class file 11 that stores a change classification class parameter that represents a change classification class that is a classification of a change selection object, and a change classification class-by-class comparison with the configuration of the first embodiment. Change definition parameter classification database 12 for storing change definition parameters as change definition sample data, and change definition sample data stored in change definition parameter classification database 12 as a first input and stored in change classification class file 11 A change definition parameter determination unit 15 is further provided that uses the change classification class parameter as a second input, determines a change definition parameter of the change selection object, and outputs the change definition parameter to the change definition parameter file 8.

  The change definition parameter determination unit 15 receives the change classification class parameters stored in the change classification class file and the change definition parameters stored in the change definition parameter classification database 12 as input, determines the change definition parameters, Output to the definition parameter file 8.

As the change classification class parameter of the selection target object, a general name of the target object such as a lake or a lake is designated, and the change inputted in the past is stored in the change definition parameter classification database 12 in advance for each class. The change definition parameter is automatically determined using the change definition sample data that is the definition parameter. For example, in the case of a lake, “area Am ² (square meter) or more (A is a preset constant), random motion”, and in the case of coral, “straight / curved shape, the area has a large aspect ratio and the same motion Change definition sample data such as “repeat pattern” is set as a change definition parameter.

  When there are a plurality of change definition sample parameters, a method of taking an average value, a maximum value, and a minimum value of change definition sample data, and a method of selecting a definition having a large number of samples are conceivable. Specifically, when it is necessary to define the area range, a method for calculating the average value, maximum value, or minimum value of the upper limit or lower limit of the area and using it as a change definition parameter, or a unique exercise method is used. If there is a need to decide, a motion method with a large number of samples is selected from among a plurality of sample motion methods. If a plurality of motion methods can be specified, a plurality of motion methods are directly designated as change definition parameters.

  As described above, according to the fifth embodiment, the user can input the change classification class name of the selection target object, and the change definition parameter is automatically set, so that it is set based on many cases. It is possible to sort with the selected parameters.

Embodiment 6 FIG. (Change definition parameter adjustment)
In the fifth embodiment, the example in which the change definition parameter is automatically determined has been shown. However, in the present embodiment, after the change definition parameter automatically determined is provisionally determined, the change definition parameter is presented to the user, and according to the user's adjustment instruction, An example of changing the change definition parameter is shown.

  FIG. 12 is a block diagram showing an image change extraction apparatus according to Embodiment 6 of the present invention. In FIG. 12, the same parts as those shown in FIG. In the sixth embodiment shown in FIG. 12, a change classification class file 11 that stores a change classification class parameter that represents a change classification class that is a classification of a change selection object, and a change classification class-by-class comparison with the configuration of the first embodiment. Change definition parameter classification database 12 for storing change definition parameters as change definition sample data, change definition sample data stored in change definition parameter classification database 12, and change classification class stored in change classification class file The parameter is input, the change definition parameter is tentatively determined, the change definition parameter tentative determination unit 16 outputs the change definition parameter tentative value, and the change definition parameter tentative value is input from the change definition parameter tentative determination unit 16 To the user as a change definition parameter The order value is adjusted based on the user instruction to determine changes defined parameter change value, further comprising a change definition parameter adjustment unit 17 outputs the change definition parameter file 8.

  The change definition parameter tentative determination unit 16 receives the change classification class parameters stored in the change classification class file and the change definition sample data stored in the change definition parameter classification database 12 as input, and temporarily determines the change definition parameters. Then, it is output as a change definition parameter provisionally determined value.

  Next, the change adjustment unit 17 inputs the change definition parameter tentative value output from the change definition parameter tentative determination unit, presents it to the user, adjusts the change definition parameter tentative value in accordance with the user instruction, The definition parameter change value is output to the change definition parameter file 8.

  An example in which “river” is designated as the change classification class is shown. Consider a case where the width of a river to be extracted as an analysis target is extremely large. The user adjusts the aspect ratio of the river provisionally determined as the change definition parameter, sets a new aspect ratio as the change definition parameter, and performs change selection processing.

  As described above, according to the sixth embodiment, by allowing the user to adjust automatically determined change definition parameters, exceptionally wide rivers or exceptionally small lakes can be extracted. Cases with change definition parameter values can also be handled.

  Further, the present invention is not limited to the above-described embodiments, and it is needless to say that all possible combinations of these embodiments are included.

  1 Coherence map file, 2 noise removal unit, 3 coherence average calculation unit, 4 change extraction unit, 5 change time adjustment unit, 6 feature quantity extraction unit, 7 change selection unit, 8 change definition parameter file, 9 change selection data file, DESCRIPTION OF SYMBOLS 10 Ground truth information file, 11 Change classification class file, 12 Change definition parameter classification database, 13 Change definition parameter classification storage part, 14 Change classification class update part, 15 Change definition parameter determination part, 16 Change definition parameter provisional determination part, 17 Change definition parameter adjustment unit.

Claims (6)

A coherence map file storing N-1 coherence maps, which are distributions of coherence between two images, based on images obtained by N observations at different times for the same region;
The coherence map that is the complex correlation distribution between the two images stored in the coherence map file is input, noise is removed from the coherence map, and noise-removed image data that is the image after noise removal is output. A noise removal unit to perform,
Coherence map group that is a collection of coherence maps between two images stored in the coherence map file is used as input, and temporal and spatial average values of pixels constituting the coherence map are calculated, and as coherence average data A coherence average calculator for outputting to a coherence map file;
Using a coherence map between two images stored in the coherence map file and coherence average data calculated from a plurality of coherence maps as input, using the coherence average data and time-series change information of the coherence map The change occurrence time data indicating the change occurrence time of the change occurrence area, the change continuation status data that is information as to whether or not the change occurrence has occurred steadily, and the judgment value information used for the change extraction A change extraction unit that outputs change extraction data composed of upper and lower limit information data and change extraction pixel data representing a change extraction region;
The change extraction data output from the change extraction unit is input after the time adjustment is performed by detecting the shift of the change occurrence time from the change occurrence time data, the change continuation status data, and the upper / lower limit information data. A change time adjustment unit that outputs change extraction data after time adjustment; and
A change definition parameter file that stores change definition parameters that are parameters related to a selection target change method and a selection target change area;
The noise removal image data output from the noise removal unit, the time-adjusted change extraction data output from the change time adjustment unit, and the change definition parameter stored in the change definition parameter file are input, and the change A feature amount extraction unit that determines a feature amount extraction method based on the definition parameter, performs feature amount extraction, outputs feature amount extraction data that is a feature amount extraction result, time-adjusted change extraction data, and change definition parameters; ,
The feature quantity extraction data, the time-adjusted change extraction data and the change definition parameter output from the feature quantity extraction unit are input, and a change selection method is determined based on the feature quantity extraction data and the change definition parameter, and feature quantity extraction is performed. The change is selected from the data and the time-adjusted change extraction data, the change occurrence time data indicating the change occurrence time, the change continuation status data indicating the time series continuation status of the change, and the change selection area and the change selection area. An image change extraction apparatus comprising: a change selection unit that stores change selection data including change selection pixel data including an ID that is uniquely expressed in a change selection data file as an overall output. The image change extraction device according to claim 1,
A ground truth information file for storing ground truth information is further provided.
The change selection unit, as pre-processing for change selection, removes an unselected change based on ground truth information that is terrain information of an imaging region stored in the ground truth information file. Extraction device. The image change extraction device according to claim 1,
A change classification class file that stores parameters representing a change classification class that is a classification of a change selection object;
The change classification parameter stored in the change classification class file is a first input, the change definition parameter stored in the change definition parameter file 8 is a second input, and the change definition parameter for each change classification class And a change definition parameter classification storage unit for storing in a change definition parameter classification database as change definition sample data which is a change definition parameter for each change classification class. The image change extraction device according to claim 1,
A change definition parameter classification database for storing change definition sample data for each change classification class;
The change definition parameter stored in the change definition parameter file is used as a first input, and the change definition sample data for each change classification class stored in the change definition parameter classification database is used as a second input. A change classification class updating unit that learns a change classification class from parameters, determines a new change classification class, updates a class classification of the change definition parameter classification database, and outputs to the change definition parameter classification database; A featured image change extraction device. The image change extraction device according to claim 1,
A change classification class file that stores a change classification class parameter representing a change classification class that is a classification of a change selection object;
Change definition parameter classification database that stores change definition class-specific change definition parameters as change definition sample data;
Change definition sample data stored in the change definition parameter classification database as a first input, change classification class parameter stored in the change classification class file as a second input, and change of a change selection object An image change extraction apparatus further comprising: a change definition parameter determination unit that determines a definition parameter and outputs the change parameter to the change definition parameter file. The image change extraction device according to claim 1,
A change classification class file that stores a change classification class parameter representing a change classification class that is a classification of a change selection object;
Change definition parameter classification database that stores change definition class-specific change definition parameters as change definition sample data;
Using the change definition sample data stored in the change definition parameter classification database and the change classification class parameter stored in the change classification class file as input, the change definition parameter is provisionally determined, and the change definition parameter is determined. A change definition parameter tentative determination part to output as a tentative value;
Presenting the change definition parameter tentative value from the change definition parameter tentative determination unit to the user as an input, and adjusting the presented change definition parameter tentative value based on the user's instruction to determine the change definition parameter change value; An image change extraction apparatus, further comprising: a change definition parameter adjustment unit that outputs the change definition parameter file.

Images