Learning data generation method and learning data generation device

Abstract

To provide a learning data generation method and a learning data generation device that can efficiently generate learning data for causing artificial intelligence (AI) extracting various types of symbols from a drawing to learn the various types of symbols.SOLUTION: Provided is a learning data generation method for generating learning data for causing artificial intelligence (AI) extracting symbols from a drawing to learn various types of symbols to extract, and the method executes a symbol information acquisition step, an extracted image forming step, a symbol area setting step, a symbol image information forming step, and an annotation data forming step of forming annotation data in which symbol image information and symbol information of symbols included in a symbol image in the symbol image information are associated with each other. Since the annotation data, which is learning data for causing the artificial intelligence (AI) to learn various types of symbols to extract from symbols included in the drawing, can be automatically generated, learning data having a large number of pieces of symbol image information and symbol information can be easily formed.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a learning data creation method and a learning data creation device, and more particularly to a learning data creation method and a learning data creation device for creating learning data used for learning in order to have an artificial intelligence (AI) that extracts various symbols and characters written near the symbols that supplement the specifications of the facilities, equipment, etc. represented by the symbols from various drawings. [Background technology]

[0002] Currently, CAD software is commonly used to create various drawings. When drawings created using CAD software are provided, they are sometimes provided not as CAD data but as PDF data or paper-based drawings (hereinafter simply referred to as "paper-based drawings, etc."), and it may be necessary to obtain various information contained in these drawings. For example, architectural design drawings contain a large amount of information such as equipment symbols (hereinafter simply referred to as "symbols"), side notes related to the symbols (hereinafter simply referred to as "side notes"), and wiring, and it may be necessary to extract the symbols and side notes. Furthermore, because symbols and side notes are written in association with each other, it is necessary to accurately extract them while keeping them associated, or to accurately associate symbols and side notes extracted separately.

[0003] For example, Patent Documents 1 and 2 disclose a technique for extracting symbols from a handwritten drawing and annotating the extracted symbol image with symbol information based on information for annotating the symbol stored in a database. In recent years, technology has also been developed that uses artificial intelligence (AI) to extract symbols and annotations in order to make the process more efficient. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 08-083333 [Patent Document 2] Japanese Patent Application Publication No. 08-083334 Summary of the Invention [Problem to be solved by the invention]

[0005] Incidentally, when using artificial intelligence (AI) to extract symbols and annotations, in order to improve accuracy, it is necessary to have the AI ​​learn a large number of symbols and annotations, and the more symbols and annotations that are learned, the more accuracy can be expected to improve.

[0006] However, currently, training data for learning symbols and annotations is created by workers using existing tools to extract symbols and annotations from original drawings and then adding information to the extracted symbols and annotations. With this method, the more symbols and annotations there are, the more effective the training data becomes, but the greater the burden on the worker. If training data could be created automatically from original drawings, it would be possible to create training data that could reduce the burden on workers and improve the accuracy of symbol and annotation extraction using artificial intelligence (AI).

[0007] In view of the above circumstances, the present invention aims to provide a training data creation method and a training data creation device that can appropriately and efficiently create training data for artificial intelligence (AI) that extracts various symbols from drawings and learns various symbols. [Means for solving the problem]

[0008] The training data creation method of the first invention is a training data creation method for creating training data for learning various symbols to be extracted by an artificial intelligence (AI) that extracts symbols from drawings, and is characterized by carrying out the following steps: a symbol information acquisition step for acquiring symbol information related to symbols from original drawing data containing information about symbols; an extraction image formation step for forming an extraction image that is a raster image from the original drawing data; a symbol area setting step for setting a symbol area including the symbol in the extraction image based on the symbol information acquired in the symbol information acquisition step; a symbol image information formation step for extracting a symbol image corresponding to the symbol area from the extraction image based on the symbol area set in the symbol area setting step to form symbol image information; and an annotation data formation step for forming annotation data that associates the symbol image information formed in the symbol image information formation step with the symbol information of a symbol included in the symbol image of the symbol image information. A second aspect of the present invention is a learning data creation method according to the first aspect of the present invention, wherein the sign includes a symbol and an annotative note related to the symbol, and the symbol information acquisition step acquires symbol information related to the symbol and annotative note information related to the annotative note, respectively, and the symbol area setting step sets a symbol area including the symbol in the extraction image and an annotative note area including the annotative note in the extraction image based on the symbol information and the annotative note information acquired in the symbol information acquisition step, respectively, and the symbol image information formation step sets the extraction image information set in the symbol area setting step. The method is characterized in that a symbol image corresponding to the symbol area and a side note image corresponding to the side note area are extracted from the symbol image information and side note image information are formed, respectively, and in the annotation data formation process, symbol annotation data is formed by associating the symbol information of the symbol included in the symbol image information formed in the symbol image information formation process with the symbol image information, and side note annotation data is formed by associating the side note information of the side note included in the side note image information formed in the symbol image information formation process with the side note image information. <Learning data creation device> The learning data creation device of the third invention is a learning data creation device that creates learning data for learning various symbols to be extracted by an artificial intelligence (AI) that extracts symbols from drawings, and is characterized by comprising: a symbol information acquisition unit that acquires symbol information related to symbols from original drawing data that contains information about symbols; an extraction image formation unit that forms an extraction image that is a raster image from the original drawing data; a symbol area setting unit that sets a symbol area including the symbol in the extraction image based on the symbol information acquired by the symbol information acquisition unit; a symbol image information formation unit that extracts a symbol image corresponding to the symbol area from the extraction image based on the symbol area set by the symbol area setting unit and forms symbol image information; and an annotation data formation unit that forms annotation data that associates the symbol image information formed by the symbol image information formation unit with the symbol information of a symbol included in the symbol image of the symbol image information. A fourth aspect of the present invention is a learning data creation device according to the third aspect of the present invention, wherein the sign includes a symbol and an annotative note related to the symbol, the symbol information acquisition unit has a function of acquiring symbol information related to the symbol and annotative note information related to the annotative note, respectively, the symbol area setting unit has a function of setting a symbol area including the symbol in the extraction image and an annotative note area including the annotative note in the extraction image, based on the symbol information and the annotative note information acquired by the symbol area setting unit, respectively, and the symbol image information formation unit forms a symbol image from the extraction image set by the symbol area setting unit. The annotation data forming unit has a function of extracting a symbol image corresponding to the symbol area and a side note image corresponding to the side note area to form symbol image information and side note image information, respectively, and the annotation data forming unit has a function of forming symbol annotation data that associates the symbol information of the symbol included in the symbol image information formed by the symbol image information forming unit with the symbol image information, and side note annotation data that associates the side note information of the side note included in the side note image information formed in the symbol image information forming step with the side note image information. [Effects of the Invention]

[0009] <How to create learning data> According to the first invention, annotation data can be automatically created as learning data for artificial intelligence (AI) to learn various symbols to be extracted from symbols contained in drawings, making it possible to easily create learning data having a large number of symbol image information and symbol information. According to the second invention, artificial intelligence (AI) can learn symbols so that symbols and annotations can be appropriately extracted even when the layout (position of the annotation) or size (relative size of the two) of the symbols and annotations differs in the target drawing data from which the symbols are to be extracted. <Learning data creation device> According to the third invention, annotation data can be automatically created as learning data for artificial intelligence (AI) to learn various symbols to be extracted from symbols contained in drawings, making it possible to easily create learning data having a large number of symbol image information and symbol information. According to the fourth invention, artificial intelligence (AI) can learn symbols so that symbols and annotations can be appropriately extracted even when the layout (position of annotation) or size (relative size of the two) of the symbols and annotations differs in the target drawing data from which the symbols are to be extracted. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 2 is a flowchart of a learning data creation method according to the present embodiment. [Figure 2] 1 is a schematic block diagram of a learning data creation device 1 according to the present embodiment. [Figure 3] (A) is a diagram showing an example of an original drawing file S1, (B) is a diagram showing an example of setting a symbol area CA including a symbol area SA and a side note area AA in an extraction image file S2, and (C) is an explanatory diagram enlarging the symbol area SA and side note area AA set in Figure 3(B). [Figure 4] (A) is an example of a symbol information list of the symbol information list, and (B) is an example of a side information list of the symbol information list. [Figure 5](A) is an example of a symbol image information list of a symbol image information list, (B) is an example of an ancillary image information list of a symbol image information list, (C) is an example of a symbol area information list of a symbol area information list, and (D) is an example of an ancillary area information list of a symbol area information list. [Figure 6] (A) is an example of a symbol annotation data list in the annotation data list, and (B) is an example of a side note annotation data list in the annotation data list. DETAILED DESCRIPTION OF THE INVENTION

[0011] The learning data creation method of this embodiment is a method for creating learning data for learning by an artificial intelligence (AI) that extracts symbols, footnotes, and the like from drawing data such as construction drawings.

[0012] The drawing file from which learning data is created by the learning data creation method of this embodiment is a file containing data on the shapes and attribute information of symbols and other elements depicted on the drawing. For example, a drawing file (vector format drawing file) formed from so-called BIM data, such as equipment CAD data, IFC data, or BE-Bridge data, can be used as the drawing from which learning data is created. Note that, hereinafter, the drawing file from which learning data is created will simply be referred to as the original drawing file.

[0013] In the following description, the original drawings from which symbols and other data are extracted will be described as construction drawings; however, the drawings from which artificial intelligence (AI) extracts symbols and other data are not limited to construction drawings. In other words, the original drawings from which learning data is created using the learning data creation method of this embodiment may be various drawings other than construction drawings, as long as they contain various symbols, and the extracted symbols are not particularly limited. In the following description, since the original drawings from which symbols and other data are extracted are construction drawings, the term "symbol" includes "symbol" and "notes." This "symbol" refers to a symbol that diagrammatically represents an electrical or mechanical component used in the installation of a building. Furthermore, "notes" refers to letters or symbols that supplement the specifications of the electrical or mechanical component represented by the symbol and are written adjacent to the symbol.

[0014] <Learning Data Creation Device 1 of This Embodiment> First, a learning data creation device 1 that implements the learning data creation method of this embodiment will be described. As shown in Figure 2, the learning data creation device 1 may have a function to display drawings, data, etc. on a display DS, and a function IS to input necessary information or information to instruct necessary operations using input devices such as a keyboard or mouse.

[0015] <Data Entry Section 2> As shown in FIG. 2, an apparatus 1 for implementing the training data creation method of this embodiment (hereinafter simply referred to as "training data creation apparatus 1") has a data input unit 2 that inputs an original drawing file S1 (see FIG. 3(A)) from which training data is created. This data input unit 2 has a function (data receiving function) of accepting data of a vector-format drawing file, such as a vector PDF file, from a storage medium such as a USB or receiving it via a line such as the Internet. The drawing file received by the data input unit 2 is transmitted from the data input unit 2 to the storage unit 10 as the original drawing file S1 and stored therein. The original drawing file S1 contains a plurality of symbols, i.e., a plurality of symbols, and annotations written adjacent to each symbol.

[0016] <Symbol Information Acquisition Unit 3> As shown in FIG. 2, the learning data creation device 1 includes a symbol information acquisition unit 3. The symbol information acquisition unit 3 has a function of acquiring information contained in the original drawing file S1 received by the data input unit 2. The information contained in the original drawing file S1 refers to information about the symbols and annotations written in the original drawing file S1 (hereinafter referred to as symbol information and annotation information). For example, the symbol information includes, for each symbol, symbol shape information (such as the size and orientation of the symbol (direction and inclination)), symbol position information (coordinates on the drawing), and symbol attribute information (such as name and specifications) (see each line in FIG. 4(A)). Furthermore, the annotation information includes, for each annotation, annotation shape information (such as the size and orientation of the annotation (direction and inclination)), position information (coordinates on the drawing), and annotation attribute information (such as name and specifications) (see each line in FIG. 4(B)). The acquired symbol information and annotation information are transmitted from the symbol information acquisition unit 3 to the storage unit 10 and stored therein.

[0017] In the following description, both the symbol information and the side information may be collectively referred to as symbol information. Furthermore, the symbol information (symbol information and ancillary information) acquired by the symbol information acquisition unit 3 may be transmitted to and stored in the memory unit 10 as a symbol information list (see Figures 4(A) and (B)) that lists each symbol information for each symbol, or when each symbol information is transmitted from the symbol information acquisition unit 3 to the memory unit 10, a symbol information list that associates each symbol information with each symbol may be created in the memory unit 10.

[0018] <Extraction image forming unit 4> As shown in Fig. 2, the learning data creation device 1 has an extraction image formation unit 4. The extraction image formation unit 4 has a function of forming an extraction image file S2 containing an extraction image from an original drawing file S1 received by the data input unit 2. Specifically, the extraction image formation unit 4 has a function of forming an extraction image file S2 containing an extraction image that is a raster format file (a drawing file in which an image is composed of pixel data) from the original drawing file S1 that is a vector format file. The extraction image file S2 formed by the extraction image forming unit 4 is sent to the storage unit 10 and stored therein.

[0019] <Symbol area setting section 5> As shown in FIG. 2, the learning data creation device 1 has a symbol area setting unit 5. The symbol area setting unit 5 has a function of setting a symbol area CA including each symbol for all symbols described in the extraction image included in the extraction image file S2 based on the symbol information created by the symbol information acquisition unit 3 (see FIGS. 3B and 3C). Specifically, for all symbols (symbols and side notes), the symbol area CA has a function of setting a symbol area CA so that the entire symbol is included based on the symbol position information and symbol size included in the symbol information for each symbol. Information about the set symbol area CA is then transmitted from the symbol area setting unit 5 to the memory unit 10 and stored therein.

[0020] For example, if the center position of a symbol is stored as the symbol position information, the symbol area setting unit 5 sets a rectangular area that includes the entire symbol as the symbol area CA based on the center position of the symbol and the size of the symbol included in the symbol information. When a rectangular area is set as the symbol area CA, the coordinates (position information) of the X-axis and Y-axis of the diagonal corners of the rectangular symbol area CA are included (see the rows in Figures 5(C) and 5(D)). Note that, within the symbol area CA, the area corresponding to the symbol is called the symbol area SA (see the rows in Figure 5(C)), and the area corresponding to the footnote is called the footnote area AA (see the rows in Figure 5(D)). When both are expressed together, they are simply called the symbol area CA.

[0021] In addition, the information on the symbol area CA set by the symbol area setting unit 5 may be transmitted to and stored in the memory unit 10 as a symbol area information list (see Figures 5(C) and (D)) that lists the information on each symbol area CA for each symbol, or when each symbol information is transmitted from the symbol area setting unit 5 to the memory unit 10, a symbol area information list may be formed in the memory unit 10 that associates the information on each symbol area CA with each symbol. Also, as shown in Figure 4, if the coordinates of a rectangular area (position coordinates of diagonal corners in Figure 4) are included as the symbol position information, that rectangular area is set as the symbol area CA, and the coordinates of the rectangular area in the symbol position information become the X-axis and Y-axis coordinates (position information) of the diagonal corners of the rectangular symbol area CA.

[0022] <Symbol Image Information Formation Unit 6> As shown in FIG. 2, the learning data creation device 1 includes a symbol image information forming unit 6. The symbol image information forming unit 6 has a function of extracting an image of an area corresponding to the symbol area CA set by the symbol area setting unit 5 from the extraction image file S2 created by the extraction image forming unit 4. Specifically, the symbol image information forming unit 6 has a function of extracting an image (symbol image) included in an area corresponding to the symbol area CA from the extraction image included in the extraction image file S2, and forming symbol image information including this symbol image. The symbol image information is transmitted from the symbol image information forming unit 6 to the storage unit 10 and stored therein. Note that the symbol area CA is set to include each symbol, so the symbol image included in the symbol image information is an image including each symbol (see the lines in FIGS. 5(A) and (B)). Furthermore, of the symbol image information, symbol image information including a symbol is referred to as symbol image information (see the lines in FIG. 5(A)), and symbol image information including a footnote is referred to as footnote image information (see the lines in FIG. 5(B)).

[0023] The symbol image information may be transmitted from the symbol image information forming unit 6 to the memory unit 10 as a symbol image information list (see Figures 5(A) and (B)) that lists each symbol image information for each symbol, and stored therein, or when each symbol image information is transmitted from the symbol image information forming unit 6 to the memory unit 10, a symbol image information list may be formed in the memory unit 10 that associates each symbol image information with each symbol.

[0024] <Annotation Data Formation Unit 7> As shown in FIG. 2, the learning data creation device 1 has an annotation data formation unit 7. The annotation data formation unit 7 has a function of forming annotation data that associates symbol image information formed by the symbol image information formation unit 6 with symbol information. Specifically, for each symbol image information, the annotation data formation unit 7 has a function of forming annotation data that associates symbol information related to symbols included in the image of each symbol image information with the symbol image included in the symbol image information. The annotation data is transmitted from the annotation data formation unit 7 to the storage unit 10 and stored therein. Furthermore, of the annotation data, annotation data information that includes symbols is referred to as symbol annotation data information (see each line in FIG. 6(A)), and annotation data information that includes side notes is referred to as side note annotation data information (see each line in FIG. 6(B)).

[0025] The annotation data forming unit 7 may transmit and store the annotation data for each symbol as an annotation data list (see Figures 6(A) and (B)) to the memory unit 10, or when the annotation data for each symbol is transmitted from the annotation data forming unit 7 to the memory unit 10, the memory unit 10 may form an annotation data list in which each annotation data is associated with each symbol. The annotation data (and annotation data list) will serve as training data for artificial intelligence (AI) to learn which symbols to extract.

[0026] <Creating learning data> The learning data creation operation by the learning data creation device 1 described above (that is, the learning data creation method of this embodiment) will be described (see FIGS. 1 and 2).

[0027] <Symbol information acquisition process> 1 and 2, when creating learning data, the original drawing file S1 received by the data input unit 2 and stored in the memory unit 10 is received by the symbol information acquisition unit 3. Upon receiving the original drawing file S1, the symbol information acquisition unit 3 acquires symbol information from the original drawing file S1, and the acquired symbol information (see FIGS. 4(A) and (B)) is sent to and stored in the memory unit 10.

[0028] The original drawing file S1 may be received by the data input unit 2 and then directly transmitted to the symbol information acquisition unit 3. Furthermore, the symbol information may be transmitted not only to the storage unit 10 but also directly to the symbol image information forming unit 6 or the annotation data forming unit 7 . Furthermore, once the information on all symbols has been acquired, all of the symbol information (e.g., a symbol information list, etc.) may be transmitted to the memory unit 10, the symbol image information forming unit 6, or the annotation data forming unit 7, or each time symbol information is acquired for each symbol, the symbol information may be transmitted to the memory unit 10, the symbol image information forming unit 6, or the annotation data forming unit 7. Furthermore, once symbol information on a certain number of symbols has been acquired, the certain number of pieces of symbol information may be collectively transmitted to the memory unit 10, the symbol image information forming unit 6, or the annotation data forming unit 7 (e.g., as a symbol information list, etc.). In addition, when symbol information and ancillary information are acquired as symbol information, the symbol information and ancillary information are acquired and stored in the memory unit 10, or transmitted to the symbol image information formation unit 6 or the annotation data formation unit 7.

[0029] <Extraction image formation process> As shown in Figures 1 and 2, when creating learning data, an original drawing file S1 received by data input unit 2 and stored in memory unit 10 is received by extraction image formation unit 4. Upon receiving original drawing file S1, extraction image formation unit 4 forms an extraction image, which is raster format data, from original drawing file S1. Once the extraction image is formed, extraction image file S2 (see Figure 3(B)) containing the extraction image is transmitted to and stored in memory unit 10.

[0030] The original drawing file S1 may be received by the data input unit 2 and then sent directly to the extracted image forming unit 4. Furthermore, the extraction image file S2 may be sent not only to the storage unit 10 but also directly to the symbol image information forming unit 6 or the annotation data forming unit 7.

[0031] <Symbol area setting process> As shown in Figures 1 and 2, when the symbol area setting unit 5 receives symbol information and extraction image file S2 from storage unit 10 (or when symbol information is transmitted from symbol information acquisition unit 3 and extraction image file S2 is transmitted from extraction image formation unit 4), the symbol area setting unit 5 sets a symbol area CA for each symbol. That is, based on the symbol information, the symbol area setting unit 5 sets a certain area including each symbol as the symbol area CA (see Figures 3(B) and (C)). Then, information about the set symbol area CA (see Figures 5(C) and (D)) is transmitted from the symbol area setting unit 5 to storage unit 10 and stored therein.

[0032] The information on the symbol area CA may be sent not only to the storage unit 10 but also directly to the symbol image information forming unit 6 or the annotation data forming unit 7. In addition, when symbol information and ancillary information are acquired as symbol information, a symbol area SA and ancillary area AA are set as symbol areas CA, respectively, and the information of the set symbol area SA and ancillary area AA is transmitted to the memory unit 10 and stored therein, or transmitted to the symbol image information forming unit 6 or the annotation data forming unit 7. Furthermore, when the symbol areas CA of all symbols have been set, information on all symbol areas CA (for example, a symbol area information list, etc.) may be transmitted to the memory unit 10, the symbol image information forming unit 6, or the annotation data forming unit 7, or each time a symbol area CA is set for each symbol, information on the symbol area CA may be transmitted to the memory unit 10, the symbol image information forming unit 6, or the annotation data forming unit 7. Furthermore, when symbol areas CA are set for a certain number of symbols, information on the certain number of symbol areas CA may be transmitted together (for example, as a symbol area information list, etc.) to the memory unit 10, the symbol image information forming unit 6, or the annotation data forming unit 7.

[0033] <Symbol image information formation process> 1 and 2, when the symbol image information forming unit 6 receives the extraction image file S2 and the information on the symbol area CA from the storage unit 10 (or when the extraction image file S2 is sent from the extraction image forming unit 4 and the information on the symbol area CA is sent from the symbol area setting unit 5), the symbol image information forming unit 6 extracts, as a symbol image, an image of the area corresponding to the symbol area CA set by the symbol area setting unit 5 from the extraction image file S2, and forms symbol image information including this symbol image. Then, the formed symbol image information (see FIGS. 5(A) and (B)) is sent from the symbol image information forming unit 6 to the storage unit 10 and stored therein.

[0034] The symbol image information may be transmitted not only to the storage unit 10 but also directly to the annotation data forming unit 7. In addition, when a symbol area SA and an annotated area AA are set as the symbol area CA, the symbol image information and annotated image information are respectively set as the symbol image information, and the set symbol image information and annotated image information are respectively transmitted to the memory unit 10 and stored therein, or transmitted to the annotation data formation unit 7. Furthermore, when the symbol image information for all symbols is formed, all of the symbol image information (for example, a symbol image information list) may be transmitted to the storage unit 10 or the annotation data formation unit 7, or each time symbol image information is formed for each symbol, the symbol image information may be transmitted to the storage unit 10 or the annotation data formation unit 7. Furthermore, when symbol image information is set for a certain number of symbols, the certain number of symbol image information may be transmitted together (for example, as a symbol image information list) to the storage unit 10 or the annotation data formation unit 7.

[0035] <Annotation data creation process> 1 and 2, when symbol information and symbol image information are acquired from the storage unit 10 (when symbol information is transmitted from the symbol information acquisition unit 3 and symbol image information is transmitted from the symbol image information formation unit 6), the annotation data formation unit 7 creates annotation data for the corresponding symbol, which associates the symbol image included in the symbol image information with each piece of information included in the symbol information. The created annotation data (or annotation data list) is transmitted from the annotation data formation unit 7 to the storage unit 10 and stored therein. In other words, learning data (see FIGS. 6(A) and (B)) is created in which annotation data including symbol images for each symbol is listed.

[0036] As the annotation data, symbol annotation data and side note annotation data are created, and the created symbol annotation data and side note annotation data are transmitted to the storage unit 10 and stored therein. Furthermore, all annotation data (e.g., an annotation data list, etc.) may be transmitted to the storage unit 10 once annotation data for all symbols have been formed, or annotation data may be transmitted to the storage unit 10 each time annotation data for each symbol is formed. Furthermore, once annotation data has been set for a certain number of symbols, a certain number of annotation data may be transmitted collectively (e.g., as an annotation data list, etc.) to the storage unit 10.

[0037] As described above, in the learning data creation method of this embodiment, the original drawing file S1 contains information about symbols, including their size and position. Therefore, based on this original drawing file S1, the creation of symbol images (i.e., learning image data) and annotation data can be automated. This facilitates the creation of annotation data (i.e., learning data) based on the original drawing file S1, even if the image in the original drawing file S1 contains multiple symbols (symbols and annotations), and reduces the workload of the worker creating the annotation data. Furthermore, since learning data containing multiple symbol image information and symbol information can be easily generated, the accuracy of symbol extraction by artificial intelligence (AI) trained using the learning data can be improved. Furthermore, with multiple symbol image information, various symbol images can be learned, even if unrelated wiring, characters, and other elements (noise figures) are superimposed on the drawing from which the symbols are to be extracted. This allows appropriate extraction of symbols, even if the symbols in the drawing from which the symbols are to be extracted contain noise figures.

[0038] In addition, in general, when a scale is set for design drawing data created using CAD, the size of symbols and footnotes is determined using templates provided for each scale, and information about the size of the symbols is not included in the design drawing data. For this reason, annotation data extracted from design drawing data with a set scale cannot be used to extract symbols from target drawings with different scales using artificial intelligence (AI). In other words, if the scales of the drawing from which the annotation data is extracted and the drawing from which the AI ​​extracts symbols (target extraction drawing) are different, the annotation data cannot be used as data to extract symbols from the target extraction drawing. However, if annotation data is created from an original drawing file S1 that contains information about the size of symbols, as in the learning data creation method of this embodiment, the annotation data can be used to extract symbols from object extraction drawings of different scales, even if the scale of the CAD design drawing data used to create the annotation data differs from the scale of the drawing from which the artificial intelligence (AI) extracts symbols (the object extraction drawing). In other words, if the size of the symbols included in the annotation data is matched to the size of the symbols in the object extraction drawing, the annotation data can be used to extract symbols from object extraction drawings of different scales.

[0039] Furthermore, since symbols and annotations are extracted separately and recorded as symbol annotation data and annotation annotation data, respectively (see Figures 6(A) and (B)), the system can be trained to appropriately extract symbols and annotations even when the layout (position of the annotation) or size (relative size of the two) of the symbols and annotations differs in the drawing data to be extracted.

[0040] In the learning data creation method of this embodiment described above, the original drawing file S1 contains information about symbols, including symbol size and position information. However, even if the original drawing file S1 does not contain symbol size, learning data can be created using the learning data creation method of this embodiment. In this case, since there is no symbol size information, the target extracted drawing is limited to a drawing with the same scale as the original drawing file S1 from which the annotation data was created.

[0041] <About the symbol domain CA> It is desirable that the symbol area setting unit 5 sets the symbol area CA in the extraction image included in the extraction image file S2 so that it includes each symbol in its entirety but does not include the surrounding graphics of each symbol as much as possible. For example, as shown in Figures 3(B) and (C), when setting a symbol area SA including a symbol as the symbol area CA, it is desirable to set it so that it includes the entire symbol but does not include the surrounding graphics of the symbol as much as possible. Furthermore, as shown in Figures 3(B) and (C), when setting an annotated area AA including annotated text as the symbol area CA, it is desirable to set it so that it includes the entire annotated text but does not include the surrounding graphics of the annotated text as much as possible. [Industrial Applicability]

[0042] The training data creation method of the present invention is suitable as a method for creating training data for learning various symbols to be extracted by artificial intelligence (AI) that extracts symbols and characters from drawings containing a large number of symbols, characters, and lines, such as construction drawings. [Explanation of symbols]

[0043] 1. Learning data creation device 2 Data entry section 3. Symbol information acquisition section 4 Extraction image formation section 5 Symbol area setting section 6. Symbol Image Information Formation Department 7 Annotation Data Formation Department 10 Storage section S1 Original drawing file S2 Extraction image file CA Symbol Area SA Symbol Area AA side note area

Claims

1. A learning data creation method for creating learning data for learning various symbols to be extracted by an artificial intelligence (AI) that extracts symbols from drawings, comprising: a symbol information acquisition step of acquiring symbol information related to the symbol from original drawing data including information related to the symbol; an extraction image forming step of forming an extraction image, which is a raster image, from the original drawing data; a symbol area setting step of setting a symbol area including the symbol in the extraction image based on the symbol information acquired in the symbol information acquisition step; a symbol image information forming step of extracting a symbol image corresponding to the symbol area from the extraction image based on the symbol area set in the symbol area setting step, and forming symbol image information; and forming annotation data that associates the symbol image information formed in the symbol image information forming step with the symbol information of the symbol included in the symbol image of the symbol image information. A method for creating learning data.

2. The symbol is a symbol and a footnote associated with the symbol, In the symbol information acquisition step, Acquire symbol information relating to the symbol and side note information relating to the side note, In the symbol area setting step, Based on the symbol information and the annotated information acquired in the symbol information acquisition step, a symbol area including the symbol and an annotated area including the annotated information are respectively set in the extraction image; In the symbol image information forming step, extracting a symbol image corresponding to the symbol area and a side note image corresponding to the side note area from the extraction image set in the symbol area setting step to form symbol image information and side note image information, respectively; In the annotation data forming step, symbol annotation data that associates the symbol information of the symbol included in the symbol image information formed in the symbol image information forming step with the symbol image information, and side note annotation data that associates the side note information of the side note included in the side note image information formed in the symbol image information forming step with the side note image information, respectively.

2. The method for creating learning data according to claim 1.

3. A learning data creation device that creates learning data for learning various symbols to be extracted by an artificial intelligence (AI) that extracts symbols from drawings, a symbol information acquisition unit that acquires symbol information related to the symbol from original drawing data that includes information related to the symbol; an extraction image forming unit that forms an extraction image, which is a raster image, from the original drawing data; a symbol area setting unit that sets a symbol area including the symbol in the extraction image based on the symbol information acquired by the symbol information acquisition unit; a symbol image information forming unit that extracts a symbol image corresponding to the symbol area from the extraction image based on the symbol area set by the symbol area setting unit and forms symbol image information; an annotation data forming unit that forms annotation data that associates the symbol image information formed by the symbol image information forming unit with the symbol information of a symbol included in the symbol image of the symbol image information. A learning data creation device characterized by:

4. The symbol is a symbol and a footnote associated with the symbol, The symbol information acquisition unit The system has a function of acquiring symbol information relating to the symbol and side note information relating to the side note, The symbol area setting unit a function of setting a symbol area including the symbol in the extraction image and a side note area including the side note in the extraction image based on the symbol information and the side note information acquired by the symbol area setting unit, The symbol image information forming unit The symbol area setting unit has a function of extracting a symbol image corresponding to the symbol area and a side note image corresponding to the side note area from the extraction image set by the symbol area setting unit, and forming symbol image information and side note image information, respectively; In the annotation data forming unit, The symbol image information forming unit has a function of forming symbol annotation data that associates the symbol information of the symbol included in the symbol image information formed by the symbol image information forming unit with the symbol image information, and side note annotation data that associates the side note information of the side note included in the side note image information formed in the symbol image information forming step with the side note image information.

4. The learning data generating device according to claim 3.

Images