JP4463516B2 - Similarity search method - Google Patents

Description

  The present invention relates to a similarity search method.

As a method for observing and evaluating a specific event from map data, for example, a method described in Patent Document 1 is known. In this conventional example, map data is divided into a plurality of meshes, and feature quantities such as animal behavior characteristics are calculated for each mesh.
JP 2002-342690 A

  However, there are the following problems when searching for similar meshes using a method of grasping feature quantities in units of meshes including the conventional example described above. That is, for searching for similar parts, it is necessary to statistically determine a teacher model that is a criterion for similarity determination. According to the above-described conventional example, the generation of a teacher model is performed by classifying a mesh according to the presence or absence of a specific event, The feature amount of the mesh in which the specific event has occurred is obtained by statistical processing.

  On the other hand, when considering a specific event on geographic information, the specific event occurrence point is affected by the surrounding geographical influence as a point, so the position of the specific event occurrence point in the cell is considered. Otherwise, an accurate teacher model cannot be obtained.

  That is, as shown in FIG. 4C, the cell at which the specific event occurrence point 2 is at the center of the cell cannot be identified with the cell at the corner, and the cell indicated at the corner is indicated by the chain line. It is necessary to take feature values. On the other hand, the conventional method for statistically processing the feature values obtained by equating the two cells cannot generate a teacher model as an accurate evaluation reference even if the mesh size is appropriate.

  Further, the disadvantages of the above conventional method also occur when searching for similar points using a teacher model for the same reason.

  The present invention has been made to solve the above-described drawbacks, and an object of the present invention is to provide a similar point search method capable of accurately searching for geographical similarity.

  1 and 2 show the principle of the present invention. FIG. 1 is a flowchart of the present invention, and FIG. 2 is a functional block diagram of a similar point search system 5 that implements the present invention. When searching for similar points, first, a specific event occurrence point 2, a teacher model space shape, variable candidates, and search The target space 3 is determined as an initial condition. The specific event occurrence point 2 refers to a point where an event from which a similar point is searched for is generated. For example, in searching for a transplantable place of a geese butterfly, it refers to a current habitat confirmation point of the geese butterfly.

  The shape of the teacher model space refers to the shape and size of the teacher model in the geographic information space, which will be described later, and the position information of the specific event occurrence point 2, and is determined in consideration of the geographical influence on the event that occurs. To determine the shape of the teacher model space, for example, in the case of a living organism, the energy consumption equivalence line associated with various actions starting from a nesting site or a growing site is used. The movable amount due to is taken into account. Further, as shown in FIG. 5 (a), the teacher model can be assumed to be three-dimensional as shown in FIG. 5 (b) in addition to a planar one.

  The variable candidate is a candidate that becomes an explanatory variable of a regression equation described later, and is a geographical element that seems to have an influence on the occurrence of an event. This variable candidate may be an amount obtained by performing an appropriate calculation on the variable candidate other than a numerical value such as an altitude or a primary amount such as an air temperature that is directly held in the geographic information database 1. Since statistical information can also be used as explanatory variables in the geographic information database 1, a GIS (Geographic Information System) that enables quantitative analysis by superimposing various statistics on map information. It is desirable to use it.

  The search target space 3 is an area on the geographic information space for searching for similar points. When these initial conditions are input to the initial condition input unit 6, a teacher model generation step (steps) in the teacher model generation unit 7 is performed next. S1) is executed. In the teacher model generation step, first, the sample space 9 is set in the sample space setting unit 8 (step S11). As shown in FIG. 4A, the sample space 9 is set after the specific event occurrence point 2 given as the initial condition is placed on the geographic information space, and then given to the specific event occurrence point 2 under the initial condition. This is performed by applying the obtained spatial shape information.

  When this step is completed, a sample space 9 equal to the number of the specific event occurrence points 2 is developed on the geographic data space, and then, as shown in FIG. Is derived from the comparison space setting unit 10 (step S12). The comparison space 11 is an area having the same size and shape as the sample space 9, and is desirably arranged evenly in the entire search target space 3.

  Next, in step S <b> 13, variable candidate values are introduced into the sample space 9 and the comparison space 11. The variable candidate values are introduced by reading the amount of variable candidates existing in each sample space 9 or comparison space 11 from the geographic information database 1 via the database access unit.

  Thereafter, in step S14, multivariate analysis is performed on the variable candidates. The multivariate analysis in the multivariate analysis unit 12 determines whether or not the variable candidates are suitable for expressing the geographical features of the sample space 9 by a statistical method. Change or, if necessary, repeat the evaluation with different default settings, especially the size and shape of the space.

  If a good result is obtained in this step, the model definition unit 13 determines a teacher model (step S15). The teacher model is defined by a regression equation given by significant explanatory variables extracted by multivariate analysis in addition to the spatial shape information including the position of the specific event occurrence point 2.

  Following the teacher model generation step, a survey space setting step in the survey space setting unit 14 is executed (step S2). In the search space setting step, as shown in FIG. 6, the search target point 16 is specified in the search target space 3 by the search target point setting unit 15 (step S <b> 21). The survey space 4 is formed with reference to 16 (step S22). The search space 4 is congruent on the teacher model and the geographic data space, and the search object point 16 occupies the position occupied by the specific event occurrence point 2 in the teacher model.

  The method of specifying the search target point 16 can be determined as appropriate, and can be automatically extracted by the search target point setting unit 15 in addition to the user directly specifying the display device showing the search target space 3. In the automatic extraction of the search target points 16, the search target space 3 is set to be arranged in a matrix at a predetermined pitch in the entire search target space 3, or when a specific condition is satisfied, an additional search target between the lattice points is set. The point 16 can be set to be arranged.

  Further, if a primary feature is given to the specific event occurrence point 2 in advance, this feature can be used when the search target point 16 is specified. The primary feature refers to a feature that can be primarily determined in the geographic information space such as on a ridgeline, along a national road, in the vicinity of an intersection, and the like. For example, in FIG. The characteristic feature is that it is on the ridge line 17, and the search target point 16 is automatically arranged at a predetermined pitch along the ridge line 17 in the geographic information space.

  In addition, in FIG. 6B, the specific event occurrence point 2 is given as a positional feature that it is along the national road, and incidentally, the condition that the first intersection from the bend and the station is included is given. Indicates. In this case, the search target point 16 and the survey space 4 are arranged at a predetermined pitch (L) along the national road 18 as shown by a solid line in the figure, and further, as shown by a broken line in the figure, a bent portion and a station The first intersection from 19 is automatically extracted.

  When the survey space setting step is completed as described above, the space evaluation unit 20 executes a space evaluation step (step S3). The space evaluation step is for evaluating the similarity between each research space 4 and the teacher model, and after calling the explanatory variables in each research space 4 from the geographic information database 1 via the database access unit 22 (step S31). The objective variable calculation unit 23 calculates a regression equation (step S32), and the similarity determination unit 24 determines similarity (step S33).

  The similarity is determined by comparing the objective variable of the regression equation calculated by the explanatory variable in the research space 4 with the objective variable of the teacher model. The difference of the objective variable is small, or the objective variable is that of the teacher model. Larger ones are determined to be similar.

  When the similarity is determined in the space evaluation step in this way, a similar point extraction step is executed (step S4), registered in the similar point registration unit 25, and displayed on a display device or a printer as required. Are output from the output unit.

  The whole similarity search system 5 or the teacher model generation unit 7, the survey space setting unit 14, and the space evaluation unit 20 that constitute the similarity search system 5 can be achieved by a computer that is operated by a program that operates to provide these functions. .

  According to the present invention, since the teacher model is made in consideration of the positional relationship with the specific event occurrence point 2, it is possible to perform similarity determination based on an evaluation criterion that accurately reflects geographical features. Judgment accuracy is improved.

  In addition, since the survey space 4 is also set while maintaining the relationship with the specific event occurrence point, it is possible to improve the accuracy of evaluation.

  FIG. 3 shows an embodiment of the present invention that is applied to estimate the habitat area of Gifu butterflies and the possibility of transfer of the habitat. It is known that the existence of the kingfisher, which is a burdock butterfly, is a condition for the inhabiting of the geese butterfly. However, even if transplantation of Kang mallow is successful, the butterfly is not able to inhabit, but there are anchovy and azalea that become dense when it becomes an adult, and spatial conditions such as wind and humidity in the forest It is known that all of them can be transplanted. This embodiment can be used to determine the habitat area of these butterflies from various conditions along with the potential for transplantation of kang mallow.

  When searching for a transplantable point (similarity), first, a point where the existence of a geese butterfly is actually confirmed is registered in the initial condition input unit 6 as a specific event occurrence point 2. Further, the behavior range of the butterfly is considered to be a circular region centered on the breeding point, and the teacher model for determining whether transplantation is possible is set to a circular shape centering on the specific event occurrence point 2.

  Further, as a preparation for processing, variable candidates are examined and determined. In FIG. 5, as candidates for variables, the length of the water channel 26, the presence or absence of cannaids inside the green forest 27, the area / number of the pond 28, the area of the paddy field 29, the area of the green area 30 / the slope / absence of honey plants, The case where the length of the forest edge 31, the presence or absence of a factory 32, the length and density of the road 33, and the area of other land use land 34 are employed is shown.

  When determining the shape of the teacher model space, the behavior range of the butterfly is investigated and its size is set. From these, the teacher model space is set as a circular area having a predetermined radius, for example, 500 m, centering on the living area of the butterfly. The

  When the above initial conditions are determined, the sample space setting unit 8 sets the sample space 9 next. When setting the sample space 9, first, the Giant Butterfly habitat 2 is arranged on the geographic data space (Step S111), and then the sample space 9 is set around the Gifu Butterfly habitat 2 (Step S112). FIG. 4A shows a state where the sample space 9 is set.

  Next, in order to evaluate the sample space 9, the comparison space setting unit 10 sets the mesh inscribed in the sample space 9 in the entire search target space 3 (step S121), and then the sample space is included in each mesh. A comparison space 11 geometrically congruent with 9 is set (step S122). FIG. 4B illustrates a state where step S122 is completed.

  Thereafter, the values of the variable candidates are introduced into the sample space 9 and the comparison space 11 (step S130), and the multivariate analysis unit 12 evaluates the variable candidates. In this embodiment, the multivariate analysis is performed by applying the discrimination problem of logistic regression analysis (step S141). When the contribution rate due to the discrimination problem is low, the variable candidate to be adopted is changed, and this is repeated until the contribution rate reaches a predetermined expected value. In this step, if the contribution rate cannot be improved only by changing the variable candidates, the initial conditions such as the teacher model space shape are reviewed.

  Thus, when the contribution rate reaches a predetermined expected value (step S142), the model definition unit 13 determines the teacher model (step S150), and the teacher model setting step ends.

  Similarity search is performed using the teacher model. In order to apply the teacher model, the search space 4 is set in the search target space 3 by the search space setting unit 14. The search space 4 is set by first setting the search target point 16 in the search target space 3 by the search target point setting unit 15 (step S210), and then setting a circular area centered on the search target point 16 as the search space 4. (Step S220). When the primary feature for the habitat 2 is given, the search target point 16 is automatically extracted from the geographic information space.

  Subsequent to the above survey space setting step, a space evaluation step is executed. In the spatial evaluation, first, explanatory variables are extracted from the geographic information database 1 (step S310), and the objective variable calculation unit 23 calculates the objective variable of the regression equation (step S320).

  In this embodiment, since the regression equation is a logistic regression equation, and the objective variable indicates the probability (p) of occurrence of an event, the similarity evaluation of the research space 4 is performed so that the objective variable is larger than that of the teacher model. It is determined by whether or not. When the objective variable of the research space 4 is larger than that of the teacher model (step S330), it is registered as a similar point (step S340), and when it is smaller, another search target point 16 is extracted. Repeat the same process.

  Note that the embodiment shown in FIG. 3 is configured to end when similar points are extracted. However, in order to extract all of a plurality of similar points that meet the conditions, after step S340, If a step of investigating whether or not the undetermined search target point 16 remains in the search target space 3 is inserted, and if the undetermined search target point 16 remains, it is configured to return to step S210. Good.

  In the above, the transplantable land of the butterfly is illustrated, but other than this, for example, it is applicable to analogy of illegal dumping places, search of high profit store areas, and the like.

It is a flowchart which shows this invention. It is a functional block diagram of the present invention, It is a flowchart which shows embodiment of this invention. It is a figure which shows operation | movement of this invention, (a) is explanatory drawing which shows the state which set sample space around the specific event onset point, (b) is explanatory drawing which shows the state which formed the comparison space, (c) is a teacher It is a figure which shows the problem of the prior art example at the time of model generation. It is a figure which shows sample space, (a) is a figure which shows the case where it is set as a planar space, (b) is explanatory drawing which shows the case where it is set as a three-dimensional space. It is a figure which shows the setting procedure of a comparison space.

Explanation of symbols

1 Geographic information database 2 Specific event occurrence point 3 Search target space 4 Survey space

Claims (3)

Using a geographic information database composed of GIS formed by superimposing various statistics on map information, a search is made on the geographic information space for similarities to specific event occurrence points in the geographic information space on the geographic information database. A similarity search method for searching from within a target space,
Create a teacher model defined by regression equations geographic quantities as statistics derived from a predetermined area as explanatory variables in the geographic information space including a specific event expression points,
The objective variable is calculated by applying the regression equation to the research space congruent with the teacher model specified in the search target space in the geographic information space,
Comparing the obtained objective variable with the objective variable of the teacher model to determine the similarity, and extracting the similarities of the specific event occurrence points from the survey space with high similarity,
In creating the teacher model, logistic regression using a variety of geographical quantities in a plurality of predetermined areas on a geographical information space that includes specific event occurrence points at the same position and geometrically congruent with each other as variable candidates. There is a significant variable candidate obtained by performing multivariate analysis to which the discriminant problem of analysis is applied, changing the variable candidate adopted until the contribution rate by the discriminant problem reaches a predetermined expected value, and repeating the multivariate analysis. Determined as an explanatory variable,
A similarity point search method in which, when extracting the similar points, a corresponding point to a specific event occurrence point in a teacher model in a research space congruent with a teacher model in a search target space in a geographic information space is determined as a similar point.   The similarity search method according to claim 1, wherein the teacher model has a circular shape centered on a specific event occurrence point in a geographic information space. The similarity according to claim 1 or 2, wherein in determining the explanatory variable, the multivariate analysis is repeated by changing the size and shape of the space of the teacher model when the contribution rate cannot be improved only by changing the variable candidate. Point search method.

Images