JP2012128558A - Identification apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve identification accuracy of a target.SOLUTION: An identification apparatus includes a selection section, a learning section, an input receiving section, a feature extraction section, a feature amount calculation section, a similarity calculation section, an identification section and an output control section. The selection section performs selection processing multiple times for selecting a plurality of groups each including one or more learning samples among a plurality of learning samples resulting from classifying each of the learning samples to any one of a plurality of categories. The learning section learns an identification criterion to identify the plurality of groups on a selected-multiple-groups unit basis and generates an evaluation criterion including a plurality of identification criteria. The input receiving section receives input of an image based on a radar echo from a target. The feature extraction section extracts a feature vector representing a feature of a contour shape of the target from the image. The feature amount calculation section calculates a feature amount of the feature vector using the evaluation criterion. The similarity calculation section calculates similarity between the feature amount and an identification feature amount. The identification section identifies the target using the similarity. The output control section causes an output section to output an identification result.

Description

  Embodiments described herein relate generally to an identification device.

  Conventionally, as a technique for identifying an unknown target, an identification technique using an ISAR (Inverse Synthetic Aperture Radar) image created by processing a radar echo from the target is known.

  For example, when identifying the type of an unknown ship, the contour shape of the unknown ship is determined using the height of a bridge or a turret located above the main axis connecting the bow and stern of the unknown ship on the ISAR image. There is a method of expressing a vector and identifying an unknown ship type based on this vector.

  However, in this method, since the contour shape of an unknown ship is expressed by a vector using the height from the main axis, the contour dent may not be expressed, which may cause misclassification. For example, if a barrel protrudes from the bridge, the contour of the ship is represented by a vector that uses the height of the barrel from the main axis, so the space below the barrel is not represented, and the same height as the barrel There is a risk of being misclassified as another ship carrying the structure.

  By the way, for example, by using a histogram of combinations of co-occurrence relationships of luminance gradients, a technique for expressing a complex contour shape including a dent as a vector has been proposed.

JP 2002-6034 A JP 2010-44439 A

  However, vectors with high expressive power as described above tend to have higher dimensions. For this reason, when applied to an identification method using an image based on a radar echo from a target, overlearning is likely to occur during learning of an evaluation criterion used to convert a vector into a feature amount used for target identification. If overlearning occurs during the learning of the evaluation criteria, the accuracy of the evaluation criteria is deteriorated, so that the accuracy of target identification is also deteriorated.

  The identification device according to the embodiment includes a selection unit, a learning unit, an input reception unit, a feature extraction unit, a feature amount calculation unit, a similarity calculation unit, an identification unit, and an output control unit. The selection unit performs a selection process of selecting a plurality of groups including one or more learning samples from a learning sample storage unit that stores a plurality of learning samples in which each learning sample is classified into any of a plurality of categories. The learning unit learns an identification criterion for identifying the plurality of groups for each of the selected plurality of groups, and generates an evaluation criterion including the learned plurality of identification criteria. The input receiving unit receives an image input based on a radar echo from the target. The feature extraction unit extracts a feature vector expressing the feature of the target contour shape from the image. The feature amount calculation unit calculates the feature amount of the feature vector using the evaluation criterion. The similarity calculation unit calculates a similarity between the feature quantity and the identification feature quantity serving as a discrimination reference. The identification unit identifies the target using the similarity. The output control unit causes the output unit to output the identification result.

The block diagram which shows the structural example of the identification device of this embodiment. The figure which shows the example of the binarized image produced | generated from the ISAR image of this embodiment. Explanatory drawing of the example of the extraction method of the vector using the co-occurrence relationship of the brightness | luminance gradient direction of this embodiment. The figure which shows the example which displayed a part of brightness | luminance gradient direction of the image of this embodiment. Explanatory drawing of the example of the calculation method of the feature-value of the feature vector of this embodiment. The figure which shows the example of the identification table of this embodiment. Explanatory drawing which shows the example of the evaluation criterion production | generation method of this embodiment. The flowchart figure which shows the example of an identification process of this embodiment. The flowchart figure which shows the example of a learning process of this embodiment. The flowchart figure which shows the example of a learning sample storage process of this embodiment. The flowchart figure which shows the identification feature-value registration process example of this embodiment. The block diagram which shows the hardware structural example of the identification device of this embodiment.

  FIG. 1 is a block diagram illustrating an example of the configuration of the identification device 100 according to the present embodiment. As illustrated in FIG. 1, the identification device 100 includes an input unit 102, an input reception unit 104, a target extraction unit 106, a feature extraction unit 108, an evaluation criterion storage unit 110, a feature amount calculation unit 112, an identification A table storage unit 114, a similarity calculation unit 116, an identification unit 118, an output control unit 120, an output unit 122, a learning sample storage unit 124, a selection unit 126, and a learning unit 128 are provided.

  The input unit 102 can be realized by, for example, an image input interface. In addition, the input receiving unit 104, the target extracting unit 106, the feature extracting unit 108, the feature amount calculating unit 112, the similarity calculating unit 116, the identifying unit 118, the output control unit 120, the selecting unit 126, and the learning unit 128 are, for example, This can be realized by an existing control device such as a CPU (Central Processing Unit). The evaluation criterion storage unit 110, the identification table storage unit 114, and the learning sample storage unit 124 are, for example, magnetic such as a hard disk drive (HDD), a solid state drive (SSD), a read only memory (ROM), and a memory card. It can be realized by at least one of existing storage devices capable of storing optically, optically, or electrically. The output unit 122 can be realized by an existing display device such as a liquid crystal display.

  The input unit 102 inputs an image based on the radar echo from the target. In the present embodiment, the input unit 102 sequentially inputs time-series ISAR (Inverse Synthetic Aperture Radar) images generated in units of frames by processing radar echoes from unknown ships to be identified. To do. This ISAR image includes an unknown ship to be identified. Note that the input unit 102 may input an ISAR image in which a background that is not related to a ship (a target in a broad sense) is removed in advance. In this case, processing by the target extraction unit 106 described later is not necessary.

In the present embodiment, the case where the ISAR image input from the input unit 102 is data represented by a Ds (Ds ≧ 1) -dimensional vector and each element of the vector is a luminance value will be described as an example. For example, it is assumed that the ISAR image X input by the input unit 102 is represented by (X ₁ , X ₂ ,..., X _Ds ). However, the image input by the input unit 102 is not limited to this.

  The input receiving unit 104 receives an input of an image input from the input unit 102. In the present embodiment, the input receiving unit 104 sequentially receives input of time-series ISAR images including unknown ships to be identified from the input unit 102.

  The target extraction unit 106 extracts a target from the image received by the input reception unit 104. In the present embodiment, the target extraction unit 106 extracts an unknown ship region included in the ISAR image received by the input reception unit 104.

  For example, the target extraction unit 106 divides the ISAR image into a high luminance region and a low luminance region, and extracts a high luminance region as a marine region (target region in a broad sense). However, the target extraction method is not limited to this. For example, the target extraction unit 106 detects the ship axis of the ship region from each ISAR image, normalizes each ISAR image based on the detected ship axis, and superimposes each normalized ISAR image, thereby These regions may be extracted.

  The feature extraction unit 108 extracts a feature vector representing the feature of the target contour shape extracted by the target extraction unit 106 from the image received by the input reception unit 104. Specifically, the feature extraction unit 108 extracts a vector using a co-occurrence relationship in the luminance gradient direction as the feature vector. In the present embodiment, the feature extraction unit 108 uses a luminance gradient as a feature vector that represents the feature of the contour shape of the unknown ship region extracted by the target extraction unit 106 from the ISAR image received by the input reception unit 104. Extract a vector using the co-occurrence relationship of directions.

  FIG. 2 is a diagram illustrating an example of a binarized image generated from an ISAR image. FIG. 3 is an explanatory diagram of an example of a vector extraction method using the co-occurrence relationship in the luminance gradient direction. FIG. 4 is a diagram showing an example in which a part of the luminance gradient direction of the ship is displayed.

  For example, the feature extraction unit 108 generates a binarized image from the ISAR image in which the pixel of the ship area is 1 and the pixels other than the ship area are 0 (see FIG. 2), and the generated binarized image is generated. The area is divided into small areas (see the left figure in FIG. 3), and the following processing is performed for each divided small area. Specifically, the feature extraction unit 108 calculates the luminance gradient of each pixel in the small region, and the appearance frequency of two luminance gradient pairs having the offset for each predetermined offset from the calculated luminance gradient Is calculated (see the central diagram of FIG. 3). Then, the feature extraction unit 108 extracts the vector using the co-occurrence relationship of the luminance gradient direction, that is, the feature vector by arranging the appearance frequency for each pair of offset and luminance gradient (see the right diagram in FIG. 3) (see FIG. 3). 4). In the example shown in FIG. 4, the barrel protrudes from the bridge, but the space under the barrel can also be expressed. Note that the feature extraction unit 108 may generate an image from the ISAR image in which the brightness of the pixels in the ship region is kept as it is and the brightness of the pixels other than the ship region is set to 0, instead of the binarized image. . Even from this image, a vector using the co-occurrence relationship of the luminance gradient direction can be extracted by the same method as described above.

The evaluation criterion storage unit 110 stores evaluation criteria generated by the selection unit 126 and the learning unit 128 described later. In the present embodiment, the evaluation criterion storage unit 110 has evaluation criteria {F ₁ (x), F ₂ (x),... Having a discriminator F _i (x) (1 ≦ i ≦ N (N ≧ 2)). Store F _N (x)}.

  The feature amount calculation unit 112 calculates the feature amount of the feature vector extracted by the feature extraction unit 108 using the evaluation criterion stored in the evaluation criterion storage unit 110.

In this embodiment, each discriminator F _i (x) included in the evaluation criteria {F ₁ (x), F ₂ (x),..., F _N (x)} stored in the evaluation criteria storage unit 110. Assume that an evaluation value s _i is returned for the feature vector as shown in Equation (1). Here, the learning unit 128 to be described later, even when using any of the learning method classifiers F _{i (x)} is learned, the output of the discriminator F _{i (x),} a group to be identified Since it is expressed as the likelihood to belong, the mathematical formula (1) is satisfied.

F _i (X) = s _i (1)

In the present embodiment, the case where the evaluation value s _i is a scalar quantity will be described as an example. However, the present embodiment is not limited to this, and may be a vector having a plurality of elements, for example. Further, the contents described later can be similarly applied even when the evaluation value s _i is a vector. For example, when the classifier F _i (x) is learned by the learning unit 128 described later using a learning method for identifying three or more groups, the output of the classifier F _i (x) is the group to be identified. Is a vector having elements having likelihoods belonging to the respective categories included in.

FIG. 5 is an explanatory diagram illustrating an example of a method for calculating the feature quantity v (Y) of the feature vector Y. As shown in FIG. 5, the feature quantity calculation unit 112 uses the evaluation criteria {F ₁ (x), F ₂ (x) stored in the evaluation criteria storage unit 110 for the feature vector Y extracted by the feature extraction unit 108. , _..., is applied to F N (x)}, and obtains the evaluation value _{s i,} calculates the feature quantity v (Y) of the feature vector Y from the evaluation value _{s i} obtained. In the present embodiment, the feature quantity calculation unit 112 calculates the feature quantity v (Y) of the feature vector Y by arranging all the evaluation values s _i together as shown in FIG. 5 and Equation (2). .

v (Y) = (s ₁ , s ₂ ,..., s _N ) (2)

For example, it is assumed that the classifier F _i (x) is a function that discriminates two classes and outputs 0 or 1 as the evaluation value s _i . In this case, the feature amount calculation unit 112 sets the feature amount of the feature vector Y extracted by the feature extraction unit 108 as, for example, v (Y) = (1, 1, 0, 1, 0,..., 1). And a vector expressed as a combination of 0 and 1. When the feature quantity calculation unit 112 calculates the feature quantity of the feature vector Y ′ (Y ′ ≠ Y) extracted by the feature extraction unit 108, for example, v (Y ′) = (0, 1, 1, 1, (0,..., 0), which is also calculated as a vector expressed by a combination of 0 and 1, but the calculated feature quantity is different from the feature quantity of the feature vector Y. That is, in this example, a vector expressed by a combination of 0 and 1 is a unique feature amount that the ship included in the ISAR image has, and is an effective feature amount for identifying the ship. The same applies to the case where the classifier F _i (x) is a function that discriminates two or more classes and outputs a real number other than 0 and 1 as the evaluation value s _i .

However, the feature amount calculation method is not limited to the above-described method, and the feature amount may be calculated by using the acquired evaluation value s _i in any way.

When the evaluation value s _i is a vector, the feature quantity v (Y) of the feature vector Y is obtained by arranging the elements of the evaluation value s _i . That is, if the number of dimensions of the evaluation value s _i is d _i , the number of dimensions D of the feature quantity v (Y) of the feature vector Y is the sum of the number of dimensions d _i (D = d ₁ +... + D _N ). Incidentally, the dimension number d _i of each evaluation value s _i is, all may be the same value or may be different.

The feature amount calculation unit 112 may calculate the feature amount v (Y) as a histogram of the evaluation value s _i . Specifically, the feature amount calculation unit 112 discretizes the evaluation value s _i (for each element if necessary), and votes for the bin corresponding to the value to thereby determine the histogram feature amount v (Y). Can be calculated.

When the evaluation value s _i is a vector, the feature amount calculation unit 112 can calculate the histogram feature amount v (Y) by using a histogram with the number of dimensions d _i . When evaluation values s _i having different numbers of dimensions are mixed, such as a scalar quantity and a vector, the feature quantity calculation unit 112 calculates and calculates a histogram feature quantity for each evaluation value s _{i of the} same type. By arranging the histogram feature values as one vector, the histogram feature value v (Y) can be calculated.

Further, the feature quantity calculation unit 112 further arranges (connects) the feature quantity obtained by arranging the evaluation values s _i into one vector and the histogram feature quantity obtained by making the evaluation values s _i into a histogram as one vector. The feature value v (Y) may be calculated by Further, the feature amount calculation unit 112 may calculate the feature amount by performing nonlinear conversion on the evaluation value s _i . Further, the feature quantity calculation unit 112 may perform normalization such as setting the average to 0 or setting the square average to 1 for the calculated feature quantity.

  The identification table storage unit 114 stores an identification table in which a category is associated with each identification feature amount serving as an identification reference. Here, the category represents a type for identifying a target. When the target is a ship, the name, type, size, or the like of the ship is applicable. The combination of the identification feature quantity and the category may be registered in advance in the identification table. Further, the feature amount calculation unit 112 calculates the feature amount of the feature vector extracted from the ISAR image including the known ship, and associates the calculated feature amount with the category to which the known ship belongs as the identification feature amount. It may be registered in the identification table.

FIG. 6 is a diagram illustrating an example of the identification table. In the example shown in FIG. 6, the category A and the identification feature quantity v ′ (A) = (a ₁ , a ₂ ,..., A _N ) that are identification criteria as to whether or not it belongs to the category A are associated, The classification feature v ′ (B) = (b ₁ , b ₂ ,..., B _N ), which serves as an identification criterion for whether or not the category B belongs to the category B, is associated.

  The similarity calculation unit 116 calculates the similarity between the feature amount calculated by the feature amount calculation unit 112 and the identification feature amount stored in the identification table storage unit 114. In the present embodiment, the similarity calculation unit 116 uses the mathematical formula (3) to calculate the feature amount v (Y) calculated by the feature amount calculation unit 112 and each identification feature amount stored in the identification table storage unit 114. The degree of similarity with (identification feature amount v ′ (A), identification feature amount v ′ (B), etc.) is calculated.

  L = (α, β) / || α ||| β || (3)

  Here, Expression (3) is an expression for calculating the simple similarity between the vector α and the vector β. (Α, β) represents the inner product of the vectors α and β, and || α || and || β || represent the norms of the vectors α and β, respectively. For example, the similarity L between the feature quantity v (Y) and the identified feature quantity v ′ (A) is expressed by L = (v (Y), v ′ (A)) / || v (Y from Equation (3). ) |||| v ′ (A) ||

  Note that the similarity calculation unit 116 uses the feature value v (Y) calculated by the feature value calculation unit 112 and the identification table storage unit 114 using Equation (4) or Equation (5) for calculating the distance between vectors. The degree of similarity with each of the identification feature amounts (such as the identification feature amount v ′ (A) and the identification feature amount v ′ (B)) stored in the table may be calculated.

  The identification unit 118 identifies the target using the similarity calculated by the similarity calculation unit 116. Specifically, the identification unit 118 uses the similarity calculated by the similarity calculation unit 116 to identify an unknown target included in the image or specify a plurality of candidates to which the unknown target belongs. In the present embodiment, the identification unit 118 is predetermined from the identification table stored in the identification table storage unit 114 in descending order of the value of the similarity L calculated by the similarity calculation unit 116, that is, in descending order of similarity. Select a number of discriminating features. Then, the identification unit 118 identifies that the ship included in the ISAR image belongs to at least one of the categories corresponding to the selected predetermined number of identification feature amounts. It should be noted that the identification unit 118 includes a predetermined number of identification features in descending order of similarity among the identification feature amounts for which the similarity L calculated by the similarity calculation unit 116 exceeds the threshold from the identification table stored in the identification table storage unit 114. An identification feature amount may be selected.

  In addition, when the similarity calculation unit 116 calculates the similarity L using Equation (4) or Equation (5) for calculating the distance between vectors, the shorter the distance between vectors, that is, the similarity L The smaller the value, the higher the similarity. In this case, the identification unit 118 selects, from the identification table stored in the identification table storage unit 114, a predetermined number of identification feature amounts in ascending order of the similarity L calculated by the similarity calculation unit 116. To do. Note that the identification unit 118 has a smaller value of the similarity L among the identification features whose similarity L calculated by the similarity calculation unit 116 is smaller than the threshold from the identification table stored in the identification table storage unit 114. A predetermined number of identification feature amounts may be selected in order.

  The output control unit 120 causes the output unit 122 to output the identification result of the identification unit 118. In the present embodiment, the output control unit 120 causes the output unit 122 to output the category identified by the identifying unit 118 as belonging to at least one of the ships included in the ISAR image.

  The output unit 122 is controlled by the output control unit 120 and outputs the identification result of the identification unit 118.

  The learning sample storage unit 124 stores a plurality of learning samples in which each learning sample is classified into one of a plurality of categories. In the present embodiment, the case where the learning sample is a feature quantity of a feature vector that expresses the feature of the contour shape of the ship region will be described as an example, but the learning sample is not limited to this. The feature amount (learning sample in a broad sense) may be stored in the learning sample storage unit 124 in advance. In addition, the feature amount calculation unit 112 may calculate the feature amount of the feature vector extracted from the ISAR image including the ship, and store the calculated feature amount in the learning sample storage unit 124.

In the present embodiment, the learning sample storage unit 124 stores M (M ≧ 2) feature amounts. It is assumed that C (C ≧ 2) are provided as feature quantity categories, and M feature quantities are classified into M _h (1 ≦ h ≦ C) feature quantities for each category. In other words, the _{M = M 1 + ... + M} C. Here, as described above, the category represents the type used to identify the target. When the target is a ship, the category corresponds to the name, type, or size of the ship. In other words, if M feature quantities include a feature quantity of a ship of type α, a feature quantity of a ship of type β, and a feature quantity of a ship of type γ, a total of three types of feature quantities of the ship are included. C = 3.

  The selection unit 126 performs a selection process for selecting a plurality of groups including one or more learning samples from the learning sample storage unit 124 a plurality of times. In particular, the selection unit 126 selects the plurality of groups so that the number of learning samples or the number of samples included in each of the plurality of groups to be selected is substantially the same. Specifically, the selection unit 126 sets the plurality of groups so that the category number of the learning samples included in each of the plurality of groups to be selected or the difference between the groups of the sample numbers is within a predetermined range. select. In the present embodiment, it is assumed that the selection unit 126 randomly selects (random sampling) learning samples from the learning sample storage unit 124. Thereby, it can be expected that the categories of learning samples included in the plurality of groups selected by the selection unit 126 are different for each selection process of the selection unit 126. However, the selection method of the learning sample is not limited to this, and any selection criterion can be used as long as it can be expected that the category of the learning sample included in the plurality of groups to be selected is different for each selection process. Good.

  In the present embodiment, the selection unit 126 performs a selection process of randomly selecting K (K = 2) groups from the learning sample storage unit 124 so that each group includes one or more feature values. ≧ 2) Repeatedly. In particular, when performing the selection process, the selection unit 126 randomly selects K groups so that the number of feature categories included in each of the K groups is the same. The value of K may be 2 or more.

  For this reason, the selection unit 126 does not select feature amounts completely randomly from the learning sample storage unit 124, but performs weighting according to the number of feature amounts of the category, and the number of feature amounts between the K groups. It is also possible to adjust so as to reduce the difference between them, or to extract a certain number of feature quantities regardless of the number of feature quantities in the category. When the number of feature amounts is constant, the selection unit 126 may further simply select feature amounts randomly from categories, or unselected feature amounts that have not yet been selected as learning samples. May be preferentially selected from the category.

The learning unit 128 learns identification criteria for identifying the plurality of groups for each of the plurality of groups selected by the selection unit 126, and generates an evaluation criterion including the plurality of learned identification criteria. In the present embodiment, the learning unit 128 learns and learns a discriminator F _i (x) (1 ≦ i ≦ N) for identifying the K groups for each of the K groups selected by the selection unit 126. Evaluation criteria {F ₁ (x), F ₂ (x),..., F _N (x)} including the N discriminators are generated and stored in the evaluation criteria storage unit 110.

The evaluation criteria {F ₁ (x), F ₂ (x),..., F _N (x)} generated by the learning unit 128 are used for identification of the ship included in the ISAR image. Specifically, each of the N classifiers F _i (x) included in the evaluation criteria {F ₁ (x), F ₂ (x),..., F _N (x)} is an ISAR image to be identified. Specifically, the evaluation value for the ship included in the ISAR image is output. The evaluation value corresponds to, for example, a likelihood or a class number indicating which class the classifier belongs to in advance. Then, the plurality of output evaluation values are used as ship feature values included in the ISAR image, and are used for ship identification.

In the present embodiment, the learning unit 128 learns a discriminator F _i (x) that identifies K groups by a support vector machine. The support vector machine is a technique for constructing two classes of pattern classifiers. A separation plane that maximizes the distance from each data is obtained from the sample, and learning is performed on the basis of maximizing the margin. However, the learning method of the classifier is not limited to this, and an existing learning method can be used. For example, the learning unit 128 can use a learning technique such as boosting. Further, for example, when learning a discriminator that identifies three groups selected by the selection unit 126, the learning unit 128 uses a learning method such as a k-neighbor discriminator, a Bayes classification, or a neural network. Can do.

FIG. 7 is an explanatory diagram illustrating an example of an evaluation criterion generation method. In the example illustrated in FIG. 7, S _{1 to} S _M (M ≧ 11) indicate learning samples (features), and C = M. That is, in the example shown in FIG. 7, it is assumed that the categories of all learning samples are different. In the example shown in FIG. 7, it is assumed that K = 2. That is, in the example illustrated in FIG. 7, the selection unit 126 sets the number of learning samples included in each group to be 2, that is, since C = M, the number of learning samples included in each group is 2. Select two groups so that

In the example illustrated in FIG. 7, the selection unit 126 selects the group 151 a including the learning samples S ₁ and S ₅ and the group 151 _b including the learning samples S ₂ and S ₇ in the first selection process. . Further, the selecting unit 126, in the second selection process, and the group 152a including learning samples _{S 3} and _{S 8,} and the group 152b including learning samples _{S 10} and _{S M,} is selected. Further, the selecting unit 126, an N-th selection process, and the group 153a including learning samples _{S 4} and _{S 9,} and the group 153b including learning samples _{S 6} and _{S 7,} is selected.

In addition, the learning unit 128 learns the discriminator F ₁ (x) that identifies the group 151a and the group 151b selected by the selection unit 126 in the first selection process. In addition, the learning unit 128 has learned the discriminator F ₂ (x) that identifies the group 152a and the group 152b selected by the selection unit 126 in the second selection process. In addition, the learning unit 128 learns a discriminator F _N (x) that identifies the group 153a and the group 153b selected by the selection unit 126 in the N-th selection process. Thereby, the learning unit 128 generates an evaluation criterion {F ₁ (x), F ₂ (x),..., F _N (x)} including the learned N discriminators.

  FIG. 8 is a flowchart illustrating an example of a flow of identification processing performed by the identification apparatus 100 according to the present embodiment.

  First, the input receiving unit 104 receives an input of an ISAR image including an unknown ship to be identified from the input unit 102 (step S100).

  Subsequently, the target extraction unit 106 extracts an unknown ship from the ISAR image received by the input reception unit 104 (step S102).

  Subsequently, the feature extraction unit 108 uses a feature vector (co-occurrence relationship of luminance gradient directions) expressing the characteristics of the contour shape of an unknown ship extracted by the target extraction unit 106 from the ISAR image received by the input reception unit 104. Vector) is extracted (step S104).

  Subsequently, the feature amount calculation unit 112 calculates the feature amount of the feature vector extracted by the feature extraction unit 108 using the evaluation criterion stored in the evaluation criterion storage unit 110 (step S106).

  Subsequently, the similarity calculation unit 116 calculates the similarity between the feature amount calculated by the feature amount calculation unit 112 and each identification feature amount stored in the identification table storage unit 114 (step S108).

  Subsequently, the identification unit 118 selects, from the identification table stored in the identification table storage unit 114, an identification feature amount having a higher similarity calculated by the similarity calculation unit 116, and corresponds to the selected identification feature amount. When a ship included in the ISAR image is classified into at least one of the categories to be classified (step S110).

  Subsequently, the output control unit 120 causes the output unit 122 to output the identification result of the identification unit 118, that is, the category identified by the identification unit 118 that the ship included in the ISAR image is classified into at least one (Step). S112).

  FIG. 9 is a flowchart illustrating an example of a flow of a learning process performed by the identification device 100 according to the present embodiment.

  First, the selection unit 126 selects two groups each including one or more feature amounts from the learning sample storage unit 124 (step S200). At this time, the selection unit 126 selects the two groups so that the difference between the groups in the number of categories of feature amounts included in each of the two groups to be selected falls within a predetermined range.

  Subsequently, the learning unit 128 learns an identification criterion for identifying the two groups selected by the selection unit 126 (step S202).

  Subsequently, for example, if the selection has not been performed N times, the selection unit 126 determines to continue the selection (Yes in step S204), returns to step S200, and if the selection has been performed N times. Determines that the selection is not continued (No in step S204), and proceeds to step S206.

  Subsequently, the learning unit 128 generates an evaluation criterion including a plurality of identification criteria learned in step S202, and stores the evaluation criterion in the evaluation criterion storage unit 110 (step S206).

  FIG. 10 is a flowchart illustrating an example of a flow of a learning sample storage process performed by the identification apparatus 100 according to the present embodiment. The learning sample storage process may be performed during the identification process of FIG. 8 or may be performed separately from the identification process of FIG.

  First, the processing from step S300 to S306 is the same as the processing from step S100 to S106 in the flowchart of FIG.

  In step S307, the feature amount calculation unit 112 stores the calculated feature amount in the learning sample storage unit 124 as a learning sample.

  FIG. 11 is a flowchart illustrating an example of a flow of identification feature amount registration processing performed by the identification device 100 according to the present embodiment.

  First, the input receiving unit 104 receives an input of an ISAR image including a known ship from the input unit 102 (step S400).

  Subsequently, the target extraction unit 106 extracts a known ship from the ISAR image received by the input reception unit 104 (step S402).

  Subsequently, the feature extraction unit 108 includes a feature vector (co-occurrence relationship of luminance gradient directions) expressing features of the contour shape of a known ship extracted by the target extraction unit 106 from the ISAR image received by the input reception unit 104. Vector) is extracted (step S404).

  Subsequently, the feature amount calculation unit 112 calculates the feature amount of the feature vector extracted by the feature extraction unit 108 using the evaluation criterion stored in the evaluation criterion storage unit 110 (step S406).

  Subsequently, the feature amount calculation unit 112 registers the calculated feature amount in the identification table in association with the category as the identification feature amount of the category to which the known ship belongs (step S407).

  FIG. 12 is a block diagram illustrating an example of a hardware configuration of the identification device 100 according to the present embodiment. As shown in FIG. 12, the identification apparatus 100 of this embodiment includes a CPU 902, a ROM 904 that stores an identification program for identifying an unknown ship included in an ISAR image, a RAM 906, an evaluation criterion, an identification table, and the like. 908, an I / F 910 that is an interface with the HDD 908, an I / F 912 that is an interface for ISAR image input, an input device 914 such as a mouse or a keyboard, and an I / that is an interface with the input device 914 F916, a display device 918 such as a display, an I / F 920 that is an interface with the display device 918, and a bus 922 are provided, and a hardware configuration using a normal computer is provided. Note that the CPU 902, the ROM 904, the RAM 906, the I / F 910, the I / F 912, the I / F 916, and the I / F 920 are connected to each other via a bus 922.

  In the identification device 100, the CPU 902 reads out an identification program from the ROM 904 onto the RAM 906 and executes it, whereby the above-described units (input reception unit, target extraction unit, feature extraction unit, feature amount calculation unit, similarity calculation unit, identification unit) The output control unit is realized on a computer, and an unknown ship included in the ISAR image input from the I / F 912 is identified using an evaluation standard or an identification table stored in the HDD 908.

  Note that the identification program may be stored in the HDD 908. The identification program is a file in an installable or executable format and is provided by being stored in a computer-readable storage medium such as a CD-ROM, CD-R, memory card, DVD, or flexible disk (FD). You may be made to do. Further, the identification program may be provided by being stored on a computer connected to a network such as the Internet and downloaded via the network. Further, the identification program may be provided or distributed via a network such as the Internet. Further, the evaluation criteria and the identification table may be stored in the ROM 904. Further, the ISAR image may be stored in the HDD 908 and the ISAR image may be input from the I / F 910.

  As described above, in the identification device 100 of the present embodiment, for each of a plurality of groups, an identification criterion for identifying the plurality of groups, that is, a category of learning samples or a combination of categories included in each group is identified between the groups. Learn identification criteria. In the identification apparatus 100 according to the present embodiment, selection of a plurality of groups, that is, selection of a category of learning samples or a combination of categories included in each group is automatically performed by the selection unit 126. For this reason, according to the identification device 100 of the present embodiment, it is not necessary for a human to teach an identification criterion, and the cost associated with learning of the identification criterion can be reduced.

  Further, in the identification apparatus 100 of the present embodiment, the number of combinations of groups, that is, the number of combinations of learning sample categories or category combinations included in a group tends to be enormous. For this reason, according to the identification device 100 of the present embodiment, it is possible to generate the evaluation standard by learning the identification standard as much as necessary without being limited by the upper limit. Also, by using many identification criteria, it is possible to disperse misrecognition caused by overlearning and suppress the influence. In particular, in the identification device 100 of the present embodiment, a case where an identification criterion is learned using a support vector machine that is less likely to overlearn is illustrated, and in this case, the possibility of overlearning can be further reduced.

  Further, in the identification device 100 of the present embodiment, the selection unit 126 is configured so that the number of categories of learning samples included in each of the plurality of groups to be selected or the difference between the groups of the number of samples falls within a predetermined range. Randomly select the groups. That is, in the identification device 100 of this embodiment, each of the plurality of groups to be identified is selected so as to have a random and uniform distribution. The identification criterion learned to identify the plurality of groups selected in this way is determined to belong to any of the evaluation values for the target feature quantity to be identified, that is, any of the plurality of groups used for learning. Is completely random.

  Thus, for example, if the same identification criterion is used for the feature quantities of the two targets, if the two targets are the same, the value of the evaluation value is the same, that is, both targets are a plurality of identification targets. It can be expected to be similar to any of the groups. On the other hand, when the two targets are different, the value of the evaluation value is undefined, that is, uncorrelated. Here, the evaluation value for each feature quantity of two different targets is uncorrelated. For example, when normalized correlation is used for similarity calculation when performing target identification, the similarity is zero. It shows that it becomes.

  That is, in the identification device 100 of the present embodiment, the plurality of groups so that the category number of the learning samples included in each of the plurality of groups to be selected or the difference between the groups of the sample numbers is within a predetermined range. The identification criterion for identifying the plurality of groups by randomly selecting is characterized in that a conversion that separates similarities of different targets is incorporated.

  Therefore, according to the identification apparatus 100 of the present embodiment, an evaluation criterion that is effective for identifying a target included in an image, that is, including an identification criterion suitable for identification accuracy of target identification is efficiently and quickly obtained from a learning sample. Can learn.

  In the identification apparatus 100 according to the present embodiment, using such an evaluation criterion, a feature vector representing a feature of an unknown ship contour shape included in an ISAR image is converted into a feature amount. Even if the vector is a high-dimensional vector such as a vector using the direction co-occurrence relationship, it can be converted into a highly accurate feature amount, and the target identification accuracy can be increased.

  Further, since the identification apparatus 100 of the present embodiment can learn the evaluation criteria efficiently and at high speed, the evaluation criteria are recursively learned by further using the feature amount calculated using the evaluation criteria as a learning sample. This can be easily realized, and the performance of the evaluation standard can be further enhanced.

  In particular, in this embodiment, the identification criterion is learned using a support vector machine in which the time required for learning is determined according to the number of samples rather than the dimension of the vector, so that the evaluation criterion can be learned at a higher speed. it can.

  As described above, according to the present embodiment, the target identification accuracy can be increased.

  It should be noted that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

  For example, as long as each step in the flowchart of the above embodiment is not contrary to its nature, the execution order may be changed, a plurality of steps may be performed simultaneously, or may be performed in a different order for each execution.

  For example, in the above-described embodiment, an example in which the identification apparatus performs both the identification process and the learning process has been described. However, the identification process and the learning process may be performed by separate apparatuses. For example, the identification process and the learning process may be performed by the identification apparatus and the learning apparatus, respectively. In this case, the evaluation criteria generated by the learning device may be stored in the identification device.

  Further, for example, in the above-described embodiment, the case where the learning sample is the feature amount of the feature vector expressing the feature of the contour shape of the ship region has been described as an example. However, the learning sample is an ISAR image including the ship. There may be. In this case, the selection unit extracts the target feature vector included in the selected ISAR image after the selected ISAR image, converts the extracted feature vector into a feature amount, and the learning unit converts the converted feature amount. Use it to learn.

DESCRIPTION OF SYMBOLS 100 Identification apparatus 102 Input part 104 Input reception part 106 Target extraction part 108 Feature extraction part 110 Evaluation criteria memory | storage part 112 Feature quantity calculation part 114 Identification table memory | storage part 116 Similarity degree calculation part 118 Identification part 120 Output control part 122 Output part 124 Learning Sample storage unit 126 Selection unit 128 Learning unit 902 CPU
904 ROM
906 RAM
908 HDD
910 I / F
912 I / F
914 Input device 916 I / F
918 Display device 920 I / F
922 bus

Claims (9)

A selection unit that performs a selection process for selecting a plurality of groups each including one or more learning samples from a learning sample storage unit that stores a plurality of learning samples in which each learning sample is classified into one of a plurality of categories;
A learning unit that learns an identification criterion for identifying the plurality of groups for each of the plurality of selected groups, and generates an evaluation criterion that includes the learned plurality of identification criteria;
An input receiving unit for receiving an input of an image based on a radar echo from a target;
A feature extraction unit for extracting a feature vector expressing the feature of the target contour shape from the image;
Using the evaluation criterion, a feature amount calculation unit that calculates a feature amount of the feature vector;
A similarity calculation unit for calculating the similarity between the feature quantity and the identification feature quantity serving as a discrimination reference;
An identification unit for identifying the target using the similarity;
An output control unit for outputting an identification result to the output unit;
An identification device comprising:   The identification device according to claim 1, wherein the feature vector is a vector using a co-occurrence relationship in a luminance gradient direction.   The identification device according to claim 1, wherein the learning unit learns the identification criterion using a support vector machine. An identification table storage unit that stores an identification table in which a category is associated with each identification feature amount;
The similarity calculation unit calculates a similarity between the feature quantity and each identification feature quantity,
The identification unit selects a predetermined number of identification feature amounts in descending order of similarity, and identifies that the target belongs to at least one of the categories corresponding to the selected predetermined number of identification feature amounts. The identification device according to any one of claims 1 to 3.   The identification apparatus according to claim 1, further comprising a target extraction unit that extracts the target from the image. An input receiving unit for receiving an input of an image based on a radar echo from a target;
A feature extraction unit for extracting a feature vector expressing the feature of the target contour shape from the image;
A selection process for selecting a plurality of groups including one or more learning samples from a plurality of learning samples in which each learning sample is classified into any of a plurality of categories is performed a plurality of times, and the plurality of groups for each of the selected groups The feature quantity of the feature vector is calculated using the evaluation criterion stored in the evaluation criterion storage unit that stores the evaluation criterion including a plurality of identification criteria generated by learning the identification criterion for identifying A feature amount calculation unit;
A similarity calculation unit for calculating the similarity between the feature quantity and the identification feature quantity serving as a discrimination reference;
An identification unit for identifying the target using the similarity;
An output control unit for outputting an identification result to the output unit;
An identification device comprising:   The identification device according to claim 6, wherein the feature vector is a vector using a co-occurrence relationship in a luminance gradient direction. An identification table storage unit that stores an identification table in which a category is associated with each identification feature amount;
The similarity calculation unit calculates a similarity between the feature quantity and each identification feature quantity,
The identification unit selects a predetermined number of identification feature amounts in descending order of similarity, and identifies that the target belongs to at least one of the categories corresponding to the selected predetermined number of identification feature amounts. The identification device according to claim 6 or 7.   The identification apparatus according to claim 6, further comprising a target extraction unit that extracts the target from the image.

Images