CN111815602B - Building PDF drawing wall identification device and method based on deep learning and morphology - Google Patents
Building PDF drawing wall identification device and method based on deep learning and morphology Download PDFInfo
- Publication number
- CN111815602B CN111815602B CN202010641627.0A CN202010641627A CN111815602B CN 111815602 B CN111815602 B CN 111815602B CN 202010641627 A CN202010641627 A CN 202010641627A CN 111815602 B CN111815602 B CN 111815602B
- Authority
- CN
- China
- Prior art keywords
- wall
- information
- module
- recognition
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 18
- 238000000034 method Methods 0.000 title claims description 38
- 238000001914 filtration Methods 0.000 claims abstract description 47
- 238000009826 distribution Methods 0.000 claims abstract description 23
- 238000006243 chemical reaction Methods 0.000 claims abstract description 22
- 230000011218 segmentation Effects 0.000 claims abstract description 18
- 230000006870 function Effects 0.000 claims abstract description 6
- 238000005260 corrosion Methods 0.000 claims description 18
- 230000007797 corrosion Effects 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 15
- 238000005516 engineering process Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 7
- 238000003709 image segmentation Methods 0.000 claims description 7
- 238000005520 cutting process Methods 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 239000007787 solid Substances 0.000 claims description 4
- 230000002146 bilateral effect Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 claims description 3
- 230000004888 barrier function Effects 0.000 abstract description 2
- 238000002360 preparation method Methods 0.000 abstract description 2
- 230000002776 aggregation Effects 0.000 description 8
- 238000004220 aggregation Methods 0.000 description 8
- 238000010276 construction Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000011960 computer-aided design Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4038—Scaling the whole image or part thereof for image mosaicing, i.e. plane images composed of plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration by the use of local operators
- G06T5/30—Erosion or dilatation, e.g. thinning
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
Abstract
The invention discloses a building PDF drawing wall recognition device based on deep learning and image morphology, which comprises: the system comprises a PDF drawing conversion module, a drawing wall body pre-recognition module, a drawing information filtering module and a drawing wall body accurate recognition module, wherein the PDF drawing conversion module provides a data source conforming to a format for a wall body pre-recognition model; the drawing wall pre-recognition module is used for realizing a wall pre-recognition function, and the drawing wall pre-recognition module takes drawing segmentation blocks output by the PDF drawing conversion module as input to obtain a drawing wall position distribution probability map; the drawing information filtering module realizes the filtering of noise information in original drawing paper, data preparation is carried out for the accurate identification of a drawing wall body, and the drawing accurate identification module uses the drawing after preliminary noise filtering as input to realize the accurate identification of the position of the wall body in the drawing. The building PDF drawing wall recognition device gets through the data barrier between the image drawing and the structured electronic drawing, and provides a data base for subsequent numerous BIM applications.
Description
Technical Field
The invention relates to unstructured drawing data information extraction, which is applied to the fields of automatic drawing review, drawing reconstruction and the like, in particular to a building PDF drawing wall identification device and method based on deep learning and morphology.
Background
The construction drawing is an important guide basis in the construction process of buildings, and comprises a plurality of building information from different building fields, such as the structural design of the buildings, the pipeline layout information, the deployment of fire-fighting systems and the like. The resolution of a single-layer derived image of a large-scale building drawing is usually about 20K × 20K, and comprises a shaft network, an auxiliary line, a solid wall, a hollow wall, a label, a lead, doors and windows and other building field related components.
Before drawing software represented by AutoCAD (computer-aided design) appears, early drawing work of construction drawings is often completed manually, and in recent years, with the common application of CAD technology, electronic drawing gradually replaces traditional paper drawing and accumulates a large amount of drawing data.
Building Information Modeling (BIM) is a technology for realizing Building Information integration. The building information modeling technology can integrate information of a building in multiple dimensions and multiple fields, and is applied to multiple life cycles of design, construction, operation, maintenance and the like of the building.
In the application research related to the BIM data, the automatic drawing identification technology is always the focus of research and is the basis for the transition from the drawing data to the application of the BIM data. The objects to be identified in the drawing can be classified into aggregation objects and non-aggregation objects, and the aggregation objects comprise symbolic objects such as doors, smoke detectors, fire hydrants and the like. The basic elements of the objects are gathered in a small range in the drawing, the irrelevant information in the bounding box of the aggregation object is extremely small, and the spatial position information of the aggregation object can be approximately expressed by the bounding box. The non-aggregation objects comprise objects such as walls, pipelines and the like, and the objects cannot describe position information by using bounding boxes. In non-aggregation objects, wall body identification is a key for representing the spatial structure layout of a building and is also a difficulty in automatic identification of drawings, and the automatic identification of wall body objects in the drawings has great significance for standard examination of building drawings.
However, at present, a large number of electronic drawings in formats such as PDF and JPG belong to image-class unstructured data, wherein basic constituent elements of an object are pixel representations of basic geometric elements such as points, straight lines and arcs, and an effective method for identifying walls of such large building drawings is not available at present.
Firstly, the scale of the wall and the scale of the original drawing have large scale fluctuation range, so that the general target detection method is difficult to be used for wall identification.
Secondly, the structure of the wall body is not fixed, and the scale variable range is large, so that accurate description is difficult. In an actual drawing, a line representing a wall has high similarity with a drawing background on a texture level, and is difficult to distinguish through texture features.
On the task of drawing wall identification, the traditional image morphology method and the identification model based on deep learning have good effects on building house type drawing wall identification, but the traditional image morphology method and the identification model based on deep learning are difficult to expand to building drawings with more complex contents, and the main reasons are as follows:
(1) The space span of the basic elements forming the wall object in the drawing is too large. The scale ratio of the wall object to the whole drawing sheet in a specific direction is between 1. The general target detection method fails in the task of wall detection due to the overlarge scale proportion and the severe scale jitter.
(2) The structure of the wall object is not fixed. The structure and scale of the wall object have large variation range and are difficult to describe accurately.
(3) The non-aggregation object has high background similarity with the drawing. The boundaries of the wall and background information such as the axle net and the auxiliary line of the drawing are not clear, and the wall and the background information need to be distinguished by means of higher-level semantic information.
(4) The method for identifying the house type drawing wall based on morphology mainly acts on house type drawing drawings with single content and low information density. The information density in the building drawing is far higher than that of a house type drawing, and too much noise superposition causes the morphological method with weak resolving power to fail.
(5) The house type graph wall body identification method based on deep learning achieves wall body identification work in a house type graph by firstly identifying wall body structure connection points and then planning and reconstructing, the resolution ratio of a building drawing is far higher than that of a house type graph drawing, so that a depth model cannot directly process the whole drawing, the accuracy rate of a wall body structure key point identification model is greatly reduced, and the subsequent identification process is influenced.
Disclosure of Invention
In order to solve the technical problem, the invention provides a wall identification method for PDF drawings of large-scale complex buildings by combining a wall identification model based on deep learning and a wall identification algorithm based on image morphology, and realizes an automatic identification device for the wall of the drawings based on the wall identification method.
The invention provides a building PDF drawing wall recognition device based on deep learning and image morphology, which comprises: PDF drawing conversion module, drawing wall body is identification module in advance, drawing information filtering module, drawing wall body accurate identification mould, its characterized in that:
the PDF drawing conversion module provides a data source conforming to the format for the wall body pre-recognition model;
the drawing wall body pre-recognition module is used for realizing a wall body pre-recognition function, and the drawing wall body pre-recognition module takes drawing segmentation blocks output by the PDF drawing conversion module as input to obtain a drawing wall body position distribution probability map;
the drawing information filtering module is used for filtering noise information in original drawing paper and preparing data for accurate identification of a drawing wall, wherein the noise information is defined as pixel information in a non-wall structure in the drawing;
the drawing information filtering module extracts foreground information in original drawing paper, generates an information shielding mask code by taking a drawing wall position distribution probability map as guidance, and filters the foreground information extracted from the original drawing paper according to the information shielding mask code to obtain a preliminary noise filtering result;
the drawing wall accurate identification module takes the drawing subjected to preliminary noise filtering as input, and accurate identification of the wall position in the drawing is achieved.
Further, the PDF drawing conversion module converts the PDF drawing to be identified into a high-definition pixel drawing according to a given resolution, and retains detail information in the drawing; and converting the high-definition drawing image into a binary image, and extracting the central content of the binary image to obtain a drawing main body part.
Further, the drawing wall pre-recognition module is based on the image segmentation model and is combined with a supervision method which takes thermodynamic diagram as a training target in the key point recognition technology to obtain a wall pre-recognition model;
performing bilateral dithering on input data and target data in a training process, and improving the sensitivity of a wall pre-recognition model to the wall boundary; in the using process, for each input image of a drawing segmentation block, a probability graph of the wall position distribution is given, the wall position distribution probability graph is a single-channel image which is as large as the input image, and the value of each element represents the probability that a pixel at the corresponding position in the original drawing belongs to a wall structure;
and splicing the processing results of the plurality of drawing segmentation blocks to obtain a wall position distribution probability map of the whole drawing.
Furthermore, the drawing wall accurate identification module can realize wall closure, all walls with various representation modes in the drawing are converted into solid walls, corrosion operation iteration times are obtained by estimating the average width of all the walls in the drawing, the upper limit of the wall width is taken as I for operation, and a wall closure result is obtained, wherein I is the iteration times;
the drawing wall accurate identification module can realize over-corrosion reduction, and the volume of the wall is expanded towards the periphery due to corrosion budget when the wall is closed, so that expansion operation is performed by taking I as iteration times to offset the side effect of wall expansion;
the drawing wall accurate identification module can realize filtering of free lines, a closed line is formed after the drawing wall accurate identification module is subjected to the operation, namely the free lines are free lines, the free lines do not belong to one part of the wall, and therefore the information is eliminated through expansion operation with iteration number I;
the drawing wall accurate identification module can realize over-expansion reduction, and expansion operation eliminates free lines and reduces the thickness of the wall, so that the side effect of the expansion operation is counteracted through I-time corrosion operation.
The invention also provides an identification method of the identification device of the building PDF drawing wall based on deep learning and image morphology, which specifically comprises the following steps:
step 1, a user uploads PDF drawings to be identified or JPG and PNG format drawings derived from the PDF drawings on a front-end interface, and a wall identification process is started;
step 2, a PDF drawing conversion module starts a drawing preprocessing task, converts input data into image data, and converts the drawing from PDF to JPG if the input data is a PDF drawing;
step 3, the drawing wall pre-recognition module performs binarization conversion on the input drawing so as to ensure the numerical accuracy of the input data;
step 4, the drawing wall pre-recognition module performs center cutting on the binary image, and filters the peripheral non-information background part of the image to reduce the calculated amount of a subsequent model;
step 5, the drawing wall pre-recognition module performs image segmentation on the input drawing according to a redundant cutting mode to obtain a plurality of image blocks to be processed containing part of drawing semantic information;
step 6, a drawing wall pre-recognition module performs information enhancement and downsampling on the drawing segmentation blocks, wherein the information enhancement operation aims to prevent the phenomenon that drawing information is lost in the downsampling process, and the downsampling operation is performed to obtain the drawing segmentation blocks with moderate resolution;
step 7, processing the drawing segmentation blocks one by the drawing wall pre-recognition model to generate a wall position distribution probability diagram, and splicing the results;
step 8, generating a wall information shielding mask according to the wall position distribution probability graph and a preset truncation threshold by the drawing information filtering module;
step 9, the drawing information filtering module performs preliminary noise filtering on the binarized drawing based on the wall information shielding mask, namely only the original information of the mask mark in the original drawing is reserved, and the rest of the information in the drawing is filtered;
step 10, the drawing wall accurate identification module executes corrosion operation by using a one-way filter kernel on the basis of a preliminary noise filtering result, performs wall closing operation, and restores an over-corroded part through expansion operation;
step 11, performing expansion operation on the drawing wall accurate identification model by using a one-way filtering kernel, eliminating a trip line, and restoring an over-expansion part through corrosion operation;
step 12, filtering interference items in the recognition result by the drawing wall accurate recognition module through round calculation and threshold screening;
and step 13, superposing the wall body recognition result and the original drawing paper, transmitting the result back to the front-end interface, and providing a data export interface.
The benefits of the invention are:
(1) The invention provides a building PDF drawing wall identification method, which can identify the position information of a building wall in unstructured drawing data, keep high consistency with original drawing paper and convert the information in an image drawing into available structured information.
(2) The building PDF drawing wall body recognition device is realized, a series of functions such as data conversion, wall body probability distribution diagram generation, drawing information filtering, wall body accurate recognition and the like can be provided, the data barrier between the image drawing and the structured electronic drawing is opened, and a data base is provided for subsequent numerous BIM applications.
Drawings
FIG. 1 is a flow chart of a wall identification module according to the present invention.
Detailed Description
Embodiments of the present apparatus will be described in further detail below with reference to fig. 1.
As shown in fig. 1, the embodiment provides an apparatus for identifying a building PDF drawing wall based on deep learning and image morphology, which implements the wall identification and extraction function in a large unstructured image drawing, and the apparatus includes: the method comprises the following steps of PDF drawing conversion module, drawing wall body pre-recognition module, drawing information filtering module and drawing wall body accurate recognition module, pre-recognizing a wall body in a drawing through an image segmentation technology based on deep learning and a key point detection correlation method, and recognizing a wall body structure in the drawing through an image morphology method by combining an original drawing on the basis of a recognition result, wherein:
and the PDF drawing conversion module provides a data source conforming to the format for the wall pre-recognition model.
The PDF drawing conversion module converts the PDF drawing to be identified into a high-definition pixel drawing according to a given resolution, and retains detail information in the drawing; converting the high-definition drawing image into a binary image, and extracting the central content of the binary image to obtain a drawing main body part; the method comprises the steps of dividing a drawing main part into a plurality of image blocks to be processed, and performing image enhancement on the image blocks to be processed to ensure the integrity of image block structure information; and obtaining the drawing segmentation blocks with certain semantic information and moderate resolution through downsampling.
The method comprises the steps of dividing a drawing main part into a plurality of image blocks to be processed, and performing image enhancement on the image blocks to be processed to ensure the integrity of image block structure information; and obtaining the drawing segmentation blocks with certain semantic information and moderate resolution through downsampling.
The drawing wall pre-recognition module is used for realizing a wall pre-recognition function, and the drawing wall pre-recognition module takes drawing segmentation blocks output by the PDF drawing conversion module as input to obtain a drawing wall position distribution probability map.
The drawing wall pre-recognition module is used for obtaining a wall pre-recognition model by taking an image segmentation model as a basis and combining a supervision method which takes thermodynamic diagram as a training target in the key point recognition technology; performing bilateral jitter on input data and target data in a training process, and improving the sensitivity of the model to the wall boundary; in the using process, for each input image, a probability graph of the wall position distribution score is given, the wall position distribution probability graph is a single-channel image which is as large as the input image, and the value of each element represents the probability that the pixel at the corresponding position in the original graph belongs to the wall structure; and splicing the output results of the plurality of drawing segmentation blocks to obtain a wall position distribution probability map of the whole drawing.
The drawing information filtering module realizes the filtering of noise information in original drawing paper and performs data preparation for accurate identification of a drawing wall, wherein the noise information is defined as pixel information in a non-wall structure in the drawing.
The drawing information filtering module extracts foreground information in original drawing paper, generates an information shielding mask code by taking a drawing wall position distribution probability map as guidance, and filters the foreground information extracted from the original drawing paper according to the mask code to obtain a preliminary noise filtering result.
The drawing wall accurate identification module takes the drawing after the preliminary noise filtration as input, realizes the accurate identification of the wall position in the drawing, and the identification result has better continuity and tidiness. The continuity refers to that the connectivity of the wall in the recognition result keeps high consistency with the communication relation drawn by the original image, and the tidiness refers to that the edge part of the wall recognition result is smooth, the phenomenon of false recognition is less, and the outlier noise points are less.
The drawing wall body accurate identification module can realize wall body closing, all wall bodies with various representation modes in a drawing are converted into solid wall bodies, and the wall body closing result is obtained by estimating the iteration times of corrosion operation of the wall body width and taking the upper limit as I for operation;
the drawing wall accurate identification module can realize over-corrosion reduction, and the volume of the wall is expanded towards the periphery due to corrosion budget when the wall is closed, so that expansion operation is performed by taking I as iteration times to offset the side effect of wall expansion;
the drawing wall accurate identification module can realize free line filtration, and a closed line is not formed after the drawing wall accurate identification module is subjected to the operation, namely the free line which does not belong to a part of the wall, so that the information is eliminated through expansion operation with iteration number I;
the drawing wall accurate identification module can realize over-expansion reduction, and expansion operation eliminates the trip line and simultaneously reduces the wall thickness, so that the side effect is also counteracted through I-time corrosion operation; and (4) filtering interference items, obtaining all communication areas in the identification and the passing through by contour calculation, and filtering noise points which obviously do not belong to the wall body according to a threshold value.
The embodiment also provides an identification method of the identification device of the building PDF drawing wall based on deep learning and image morphology, which specifically comprises the following steps:
step 1, a user uploads PDF drawings to be identified or JPG and PNG format drawings derived from the PDF drawings on a front-end interface, and a wall identification process is started;
step 2, a PDF drawing conversion module starts a drawing preprocessing task, converts input data into image data, and converts the input data into a JPG image by resolution =600 if the input data is a PDF drawing;
step 3, carrying out binarization conversion on the input drawing to ensure the numerical value accuracy of the input data;
step 4, performing center cutting on the binary image, and filtering the peripheral background part without information of the image to reduce the calculation amount of a subsequent model;
step 5, carrying out image segmentation on the input drawing according to a redundant cutting mode to obtain a drawing block containing part of drawing semantic information;
step 6, performing information enhancement and downsampling on the divided drawing blocks, wherein the purpose of information enhancement operation is to prevent the drawing information from being lost in the downsampling process, and downsampling operation is performed to obtain drawing divided blocks with moderate resolution;
step 7, processing the drawing segmentation blocks one by the drawing wall pre-recognition model to generate a wall position distribution probability diagram, and splicing the results;
step 8, generating a wall information shielding mask according to the wall position distribution probability graph and a preset truncation threshold by the drawing information filtering module;
step 9, performing preliminary noise filtering on the binarized drawing based on the wall information shielding mask, namely only reserving the original information of the information shielding mask mark in the original drawing paper, and filtering the rest information in the drawing;
step 10, based on the preliminary noise filtering result, performing corrosion operation by using a one-way filtering kernel, performing wall closing operation, and restoring an over-corroded part through expansion operation;
step 11, performing expansion operation by using a one-way filter kernel to eliminate free lines, and restoring an over-expanded part by corrosion operation;
step 12, filtering interference items in the identification result through round calculation and threshold screening;
and step 13, the wall body recognition result and the original drawing paper are superposed, transmitted back to the front-end interface, and a data export interface is provided.
Although the present application has been disclosed in detail with reference to the accompanying drawings, it is to be understood that such description is merely illustrative and not restrictive of the application of the present application. The scope of the present application is defined by the appended claims and may include various modifications, adaptations, and equivalents of the subject invention without departing from the scope and spirit of the present application.
Claims (5)
1. A building PDF drawing wall recognition device based on deep learning and image morphology comprises: PDF drawing conversion module, drawing wall body is identification module in advance, drawing information filtering module, drawing wall body accurate identification module, its characterized in that:
the PDF drawing conversion module provides a data source conforming to the format for the wall body pre-recognition model;
the drawing wall pre-recognition module is used for realizing a wall pre-recognition function, and the drawing wall pre-recognition module takes drawing segmentation blocks output by the PDF drawing conversion module as input to obtain a drawing wall position distribution probability map;
the drawing information filtering module is used for filtering noise information in original drawing paper and preparing data for accurate identification of a drawing wall, wherein the noise information is defined as pixel information in a non-wall structure in the drawing;
the drawing information filtering module extracts foreground information in original drawing paper, generates an information shielding mask code by taking a drawing wall position distribution probability map as guidance, and filters the foreground information extracted from the original drawing paper according to the information shielding mask code to obtain a preliminary noise filtering result;
the drawing wall accurate identification module takes the drawing subjected to preliminary noise filtering as input, and accurate identification of the wall position in the drawing is achieved.
2. The device for identifying the building PDF drawing wall based on the deep learning and the image morphology as claimed in claim 1, wherein:
the PDF drawing conversion module converts the PDF drawing to be identified into a high-definition pixel drawing according to a given resolution, and retains detail information in the drawing; and converting the high-definition drawing image into a binary image, and extracting the central content of the binary image to obtain a drawing main body part.
3. The device for identifying the building PDF drawing wall based on the deep learning and the image morphology as claimed in claim 1, wherein:
the drawing wall pre-recognition module is used for obtaining a wall pre-recognition model by taking an image segmentation model as a basis and combining a supervision method which takes thermodynamic diagram as a training target in the key point recognition technology;
performing bilateral jitter on input data and target data in a training process, and improving the sensitivity of a wall pre-recognition model to the wall boundary; in the using process, for each input image of a drawing segmentation block, a probability graph of wall position distribution is given, the wall position distribution probability graph is a single-channel image which is as large as the input image, and the value of each element represents the probability that a corresponding position pixel in the original drawing belongs to a wall structure; and splicing the processing results of the plurality of drawing segmentation blocks to obtain a wall position distribution probability map of the whole drawing.
4. The device for identifying the building PDF drawing wall based on the deep learning and the image morphology as claimed in claim 1, wherein: the drawing wall accurate identification module can realize wall closure, all walls with various representation modes in the drawing are converted into solid walls, corrosion operation iteration times are obtained by estimating the average width of all the walls in the drawing, the upper limit of the wall width is taken as I for operation, and a wall closure result is obtained, wherein I is the iteration times;
the drawing wall accurate identification module can realize over-corrosion reduction, and the volume of the wall is expanded to the periphery due to corrosion budget when the wall is closed, so that expansion operation is performed by taking I as iteration times to offset the side effect of wall expansion;
the drawing wall accurate identification module can realize filtering of free lines, a closed line is formed after the drawing wall accurate identification module is subjected to the operation, namely the free lines are free lines, the free lines do not belong to one part of the wall, and therefore the information is eliminated through expansion operation with iteration number I;
the drawing wall accurate identification module can realize over-expansion reduction, and expansion operation eliminates the trip line and reduces the wall thickness at the same time, so that the side effect of the expansion operation is counteracted through I times of corrosion operation.
5. A recognition method of a recognition device of a building PDF drawing wall based on deep learning and image morphology specifically comprises the following steps:
step 1, a user uploads PDF drawings to be identified or JPG and PNG format drawings derived from the PDF drawings on a front-end interface, and a wall identification process is started;
step 2, starting a drawing preprocessing task by a PDF drawing conversion module, converting input data into image data, and converting the drawing into a JPG image by a PDF if the input data is a PDF drawing;
step 3, the drawing wall pre-recognition module performs binarization conversion on the input drawing to ensure the numerical accuracy of the input data;
step 4, the drawing wall pre-recognition module performs center cutting on the binary image, and filters the background part without information on the periphery of the image to reduce the calculated amount of a subsequent model;
step 5, the drawing wall pre-recognition module performs image segmentation on the input drawing according to a redundant cutting mode to obtain a plurality of image blocks to be processed containing part of drawing semantic information;
step 6, a drawing wall pre-recognition module performs information enhancement and downsampling on the drawing segmentation blocks, wherein the information enhancement operation aims to prevent the phenomenon that drawing information is lost in the downsampling process, and the downsampling operation is performed to obtain the drawing segmentation blocks with moderate resolution;
step 7, processing the drawing segmentation blocks one by the drawing wall pre-recognition model to generate a wall position distribution probability diagram, and splicing the results;
step 8, generating a wall information shielding mask according to the wall position distribution probability graph and a preset truncation threshold by the drawing information filtering module;
step 9, the drawing information filtering module performs preliminary noise filtering on the binarized drawing based on the wall information shielding mask, namely only the original information of the mask mark in the original drawing is reserved, and the rest of the information in the drawing is filtered;
step 10, the drawing wall accurate identification module executes corrosion operation by using a one-way filter kernel on the basis of a preliminary noise filtering result, performs wall closing operation, and restores an over-corroded part through expansion operation;
step 11, performing expansion operation on the drawing wall accurate identification model by using a one-way filtering kernel, eliminating a free line, and restoring an over-expansion part through corrosion operation;
step 12, filtering interference items in the recognition result by the drawing wall accurate recognition module through round calculation and threshold screening;
and step 13, the wall body recognition result and the original drawing paper are superposed, transmitted back to the front-end interface, and a data export interface is provided.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010641627.0A CN111815602B (en) | 2020-07-06 | 2020-07-06 | Building PDF drawing wall identification device and method based on deep learning and morphology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010641627.0A CN111815602B (en) | 2020-07-06 | 2020-07-06 | Building PDF drawing wall identification device and method based on deep learning and morphology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111815602A CN111815602A (en) | 2020-10-23 |
| CN111815602B true CN111815602B (en) | 2022-10-11 |
Family
ID=72842989
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010641627.0A Active CN111815602B (en) | 2020-07-06 | 2020-07-06 | Building PDF drawing wall identification device and method based on deep learning and morphology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111815602B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112712592B (en) * | 2021-03-26 | 2021-07-02 | 泰瑞数创科技(北京)有限公司 | Building three-dimensional model semantization method |
| CN113239438B (en) * | 2021-05-20 | 2022-11-25 | 中机中联工程有限公司 | Rapid point location arranging method in BIM (building information modeling) |
| CN113469086B (en) * | 2021-07-09 | 2023-06-06 | 上海智臻智能网络科技股份有限公司 | Method, device, equipment and medium for dividing area in building plan |
| CN113469087B (en) * | 2021-07-09 | 2023-05-16 | 上海智臻智能网络科技股份有限公司 | Picture frame detection method, device, equipment and medium in building drawing |
| CN113780190B (en) * | 2021-09-14 | 2022-10-11 | 万翼科技有限公司 | Method, equipment and storage medium for constructing space contour recognition and space detection model |
| CN113947529B (en) * | 2021-10-14 | 2023-01-10 | 万翼科技有限公司 | Image enhancement method, model training method, component identification method and related equipment |
| CN114517544B (en) * | 2022-04-21 | 2022-08-23 | 江苏拓信建筑科技有限公司 | Passive sandwich wall manufacturing method and device based on data processing |
| CN115393583B (en) * | 2022-07-21 | 2023-09-29 | 泰瑞数创科技(北京)股份有限公司 | Method for carrying out artificial intelligence semantic segmentation on wall |
| CN115393420A (en) * | 2022-08-29 | 2022-11-25 | 上海智臻智能网络科技股份有限公司 | Method and system for calculating space area of building drawing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971098A (en) * | 2014-05-19 | 2014-08-06 | 北京明兰网络科技有限公司 | Method for recognizing wall in house type image and method for automatically correcting length ratio of house type image |
| WO2016050083A1 (en) * | 2014-09-29 | 2016-04-07 | 苏州四方万象信息科技有限公司 | Method for converting two-dimensional image of building into three-dimensional structure |
| CN108763813A (en) * | 2018-06-08 | 2018-11-06 | 杭州群核信息技术有限公司 | It is a kind of to identify the method and apparatus for copying wall in figure based on deep learning |
| CN108804815A (en) * | 2018-06-08 | 2018-11-13 | 杭州群核信息技术有限公司 | A kind of method and apparatus assisting in identifying wall in CAD based on deep learning |
-
2020
- 2020-07-06 CN CN202010641627.0A patent/CN111815602B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971098A (en) * | 2014-05-19 | 2014-08-06 | 北京明兰网络科技有限公司 | Method for recognizing wall in house type image and method for automatically correcting length ratio of house type image |
| WO2016050083A1 (en) * | 2014-09-29 | 2016-04-07 | 苏州四方万象信息科技有限公司 | Method for converting two-dimensional image of building into three-dimensional structure |
| CN108763813A (en) * | 2018-06-08 | 2018-11-06 | 杭州群核信息技术有限公司 | It is a kind of to identify the method and apparatus for copying wall in figure based on deep learning |
| CN108804815A (en) * | 2018-06-08 | 2018-11-13 | 杭州群核信息技术有限公司 | A kind of method and apparatus assisting in identifying wall in CAD based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| Deep Floor Plan Recognition Using a Multi-Task Network With Room-Boundary-Guided Attention;Zhiliang Zeng, etc.;《2019 IEEE/CVF International Conference on Computer Vision》;20200227;全文 * |
| 利用Faster R-CNN自动识别建筑平面图功能构件的方法;马焜阳等;《测绘科学技术学报》;20191207;全文 * |
| 建筑构件智能识别方法研究;李伟青;《浙江大学学报(理学版)》;20050825(第04期);全文 * |
| 施工图设计阶段BIM模型的工程算量问题研究;姜泓列;《山东工业技术》;20170115(第02期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111815602A (en) | 2020-10-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111815602B (en) | Building PDF drawing wall identification device and method based on deep learning and morphology | |
| CN112967243B (en) | Deep learning chip packaging crack defect detection method based on YOLO | |
| CN110210475B (en) | License plate character image segmentation method based on non-binarization and edge detection | |
| CN111898507A (en) | Deep learning method for predicting earth surface coverage category of label-free remote sensing image | |
| CN110348435B (en) | Target detection method and system based on regional candidate network | |
| Chen et al. | Fault detection of insulators using second-order fully convolutional network model | |
| CN111079658A (en) | Video-based multi-target continuous behavior analysis method, system and device | |
| CN108171119B (en) | SAR image change detection method based on residual error network | |
| Lee | Recognizing hand-drawn electrical circuit symbols with attributed graph matching | |
| CN116797787B (en) | Remote sensing image semantic segmentation method based on cross-modal fusion and graph neural network | |
| CN112861575A (en) | Pedestrian structuring method, device, equipment and storage medium | |
| CN115424017B (en) | Building inner and outer contour segmentation method, device and storage medium | |
| CN114841972A (en) | Power transmission line defect identification method based on saliency map and semantic embedded feature pyramid | |
| Ortega et al. | Generating 3D city models from open LiDAR point clouds: Advancing towards smart city applications | |
| CN112884758A (en) | Defective insulator sample generation method and system based on style migration method | |
| Rajeswari et al. | Automatic road extraction based on level set, normalized cuts and mean shift methods | |
| CN111461121A (en) | Electric meter number identification method based on YO L OV3 network | |
| CN113888505B (en) | Natural scene text detection method based on semantic segmentation | |
| Li et al. | Real-time Safety Helmet-wearing Detection Based on Improved YOLOv5. | |
| CN114926734A (en) | Solid waste detection device and method based on feature aggregation and attention fusion | |
| CN114332644A (en) | Large-view-field traffic density acquisition method based on video satellite data | |
| CN115661429B (en) | System and method for identifying defects of boiler water wall pipe and storage medium | |
| Walder et al. | Neural network based methods for cloud classification on AVHRR images | |
| Li et al. | (Retracted) Feature extraction method CNDFA for target contour of coal and gangue based on multifractal | |
| CN113408651B (en) | Unsupervised three-dimensional object classification method based on local discriminant enhancement |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |