CN112487537A - Building surface multistage optimization extraction method based on full convolution neural network - Google Patents
Building surface multistage optimization extraction method based on full convolution neural network Download PDFInfo
- Publication number
- CN112487537A CN112487537A CN202011424539.1A CN202011424539A CN112487537A CN 112487537 A CN112487537 A CN 112487537A CN 202011424539 A CN202011424539 A CN 202011424539A CN 112487537 A CN112487537 A CN 112487537A
- Authority
- CN
- China
- Prior art keywords
- building
- contour
- neural network
- full convolution
- algorithm
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
Abstract
The invention discloses a building surface multistage optimization extraction method based on a full convolution neural network, which comprises the following steps: s1, preliminarily extracting the outline of the building by using a full convolution-based neural network, and then extracting the initial outline of the building on the basis of the preliminary outline of the building to perform polygon fitting treatment; s2, performing initial contour warping on the fitted contour by utilizing a best fit external rectangular junction Hausdorff distance algorithm; s3, finally, performing depth optimization on the local contour which cannot be optimized in the complex contour by using a Shi-Tomasi algorithm, and finally outputting a building contour result with the optimal optimization regularity.
Description
Technical Field
The invention relates to the technical field of intelligent control, in particular to a building surface multistage optimization extraction method based on a full convolution neural network.
Background
The building is a main ground object type in the urban environment, the AOI of the building is automatically extracted from the image map data, the coverage and accuracy of the GIS data can be effectively improved, and the method has important significance for improving urban planning and commercial marketing;
the building outline extraction is a process of identifying a building from image map data and acquiring the position and outline information of the building, and the problem of how to improve the quality and efficiency of information acquisition of the building as the most prominent ground feature type in the urban environment is solved.
Disclosure of Invention
The invention provides a building surface multistage optimization extraction method based on a full convolution neural network, which can effectively solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme: a building surface multistage optimization extraction method based on a full convolution neural network comprises the following steps:
s1, preliminarily extracting the outline of the building by using a full convolution-based neural network, and then extracting the initial outline of the building on the basis of the preliminary outline of the building to perform polygon fitting treatment;
s2, performing initial contour warping on the fitted contour by utilizing a best fit external rectangular junction Hausdorff distance algorithm;
and S3, finally, carrying out depth optimization on the local contour which cannot be optimized in the complex contour by using a Shi-Tomasi algorithm, and finally outputting the best optimization regular building contour result.
According to the above technical solution, the specific steps in S1 are to remove all FC layers of the FCNs network;
and changing feature maps with different sizes in level2 into the size of an original image by using different up-sampling magnifications, and then obtaining a prediction result through a full convolution layer.
According to the technical scheme, in the step S1, the building contour is extracted, the Douglas-Peucker algorithm is used for performing polygon fitting on the contour, the polygon fitting of the Douglas-Peucker algorithm is performed, the number of contour points is large due to the fact that sawteeth exist on the edge of the initial building contour extracted according to the classification principle, calculation of the subsequent contour points is not facilitated, and in order to reduce the calculated amount, the main shape characteristics of the contour are reserved through the polygon fitting method.
According to the technical scheme, the best-fit circumscribed rectangle in the S2 is extracted, and the circumscribed rectangle consistent with the inclination degree of the building axis is used as the best-fit circumscribed rectangle, so that the accuracy can be effectively improved.
According to the technical scheme, a Hausdorff distance algorithm is utilized in S2 to accurately measure the distance difference between the fitting outline of the building and the best fitting circumscribed rectangle, and the distance difference is used as a standard for judging whether the fitting outline is proper or not.
According to the technical scheme, the contour of the complex building is optimized based on Shi-Tomasi algorithm in S3, and according to the characteristics of right-angle corners of a right-angle building, the sharp small angles can be removed while the right-angle details of the contour are kept by utilizing corner detection and corner selection and rejection, so that the vertical corners conforming to the contour of the building are obtained.
Compared with the prior art, the invention has the beneficial effects that: the invention has scientific and reasonable structure and safe and convenient use, extracts the building by utilizing the characteristics of the building such as spectral information, shape texture and the like, performs multi-stage optimization on the complex building outline by full convolution neural network, polygonal fitting, best fitting external rectangle, Hausdorff distance algorithm and Shi-Tomasi algorithm, can effectively improve the extraction accuracy, thereby fully utilizing the network structure of the space context information under different scales, optimizing the dynamic range of the building size, reducing the network parameters, improving the image processing efficiency and finally improving the accuracy of the building outline extraction by optimizing the full convolution neural network algorithm.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic diagram of the structure of the process steps of the present invention.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.
Example (b): as shown in fig. 1, the invention provides a technical solution, and a building surface multistage optimization extraction method based on a full convolution neural network, comprising the following steps:
s1, preliminarily extracting the outline of the building by using a full convolution-based neural network, and then extracting the initial outline of the building on the basis of the preliminary outline of the building to perform polygon fitting treatment;
s2, performing initial contour warping on the fitted contour by utilizing a best fit external rectangular junction Hausdorff distance algorithm;
and S3, finally, carrying out depth optimization on the local contour which cannot be optimized in the complex contour by using a Shi-Tomasi algorithm, and finally outputting the best optimization regular building contour result.
According to the technical scheme, the outline of the building is preliminarily extracted based on the full convolution neural network, the network structure of the space context information under different scales can be fully utilized, the dynamic range of the size of the building is optimized, the network parameters are reduced by optimizing the algorithm of the full convolution neural network, and the image processing efficiency is improved;
the full convolution neural network classifies the image at pixel level, thereby solving the problem of image segmentation at semantic level, different from the classic CNN which classifies the image by using a full connection layer to obtain a feature vector with fixed length on a convolution layer, the FCN can accept an input image with any size, and an deconvolution layer is adopted to up-sample a feature map (feature map) of a last convolution base layer, so that the feature map is restored to the same size of the input image, thereby generating a prediction for each pixel, simultaneously reserving space information in the original input image, and finally classifying the pixels by using the feature map which is up-sampled at odd-even level.
According to the above technical solution, the specific steps in S1 are to firstly remove all FC layers of the FCNs network;
on one hand, redundant parameters can be greatly reduced;
on the other hand, the output image of the FC layer is only 1/32 of the original image, so that too much position information is lost, and the extraction of the building is hardly facilitated;
the feature maps with different sizes in level2 are changed into the size of the original image by using different up-sampling magnifications, and then the prediction result is obtained through the full convolution layer. Thus obviously improving the problem of gradient back propagation imbalance;
secondly, the shallow profile of a deep neural network has very precise location information but low semantics, while the deep profile is the opposite;
for the image segmentation problem, strong semantic information and accurate position information are needed, so that the high-level, middle-level and low-level feature maps are all fused to obtain a better effect.
According to the technical scheme, the building contour is extracted in S1, the Douglas-Peucker algorithm is used for performing polygon fitting on the contour, the polygon fitting of the Douglas-Peucker algorithm is performed, the number of contour points is large due to the fact that sawteeth exist on the edge of the initial building contour extracted according to the classification principle, calculation of the subsequent contour points is not facilitated, and in order to reduce the calculated amount, the main shape characteristics of the contour are reserved through the polygon fitting method.
According to the technical scheme, the Douglas-Peucker algorithm is an algorithm which approximately represents a curve as a series of points and reduces the number of the points. The description is as follows:
(1) a straight line AB is connected between the head point A and the tail point B of the curve, and the straight line is a chord of the curve;
(2) obtaining a point C with the maximum distance from the straight line segment on the curve, and calculating the distance d between the point C and the AB;
(3) comparing the distance with a preset threshold value threshold, if the distance is smaller than the threshold value threshold, taking the straight line segment as the approximation of a curve, and finishing the processing of the curve segment;
(4) if the distance is greater than the threshold value, dividing the curve into two segments of AC and BC by using C, and respectively carrying out 1-3 processing on the two segments of the credit;
(5) when all the curves are processed, the broken lines formed by all the dividing points are connected in sequence, and the broken lines can be used as the approximation of the curves.
According to the technical scheme, the best-fit external rectangle in the S2 is extracted, and the external rectangle which is consistent with the inclination degree of the axis of the building is used as the best-fit external rectangle, so that the accuracy can be effectively improved;
selecting a best fit external rectangle by using an axial evaluation method, firstly obtaining a minimum area external rectangle of each building, selecting a best fit external rectangle corresponding to the minimum area external rectangle by judging the inclination degree of the axis of the building and the minimum area external rectangle relative to a screen coordinate system, wherein the inclination degree is calculated by counting the difference value of horizontal and vertical coordinates of all adjacent two points on a polygon, when the probability that the difference value is less than 2-5 pixel values is larger, the polygon is approximately parallel to the screen coordinate axis, the inclination degree is considered to be positive, otherwise, the inclination degree is negative, judging whether the inclination degrees of the building polygon and the minimum area external rectangle are consistent according to the idea, and if the two are consistent, selecting the minimum area external rectangle; and conversely, selecting the minimum external rectangle as the best fitting external rectangle.
According to the technical scheme, a Hausdorff distance algorithm is utilized in S2 to accurately measure the distance difference between the fitting outline of the building and the best fitting circumscribed rectangle, and the distance difference is used as a standard for judging whether the fitting outline is proper;
as a standard for judging whether the fitted contour is proper, the Hausdorff distance algorithm has the basic principle that the similarity between samples is measured by calculating the distance between two groups of sample points, the one-way Hausdorff distance between each line segment on the building contour and the best fit external rectangle is calculated by equally dividing the building contour polygon and the best fit external rectangle line segment, so that the best fit contour of each equally divided segment is gradually selected, and a preliminary building contour extraction result is finally formed.
According to the technical scheme, the contour of the complex building is optimized based on the Shi-Tomasi algorithm in S3, and according to the characteristics of right-angle corners of a right-angle building, the sharp small angles can be removed while the right-angle details of the contour are kept by utilizing corner detection and performing corner selection and rejection, so that the vertical corners conforming to the contour of the building are obtained.
According to the technical scheme, the Shi-Tomasi algorithm is a corner point detection algorithm proposed by Shi and Tomasi, the working principle is that points with maximum curvature values in an image edge curve are found, local contour corner points are extracted by using the Shi-Tomasi algorithm and then are matched and sorted, then, the sizes of three-point two-line included angles are calculated and analyzed in sequence, useless corner points are removed, and finally, the reserved corner points are connected with a comprehensive regular building in sequence, and the method specifically comprises the following steps:
(1) extracting local contour corner points based on Shi-Tomasi, extracting local line segments which cannot be regulated on the basis of a preliminary optimization result, calculating gray level changes of local small windows w (x, y) after moving in all directions by using a Shi-Tomasi algorithm to detect the corner points, regarding the gray level in each direction window at a large change position as the corner points, and extracting all the corner points on the local line segments according to the change position;
(2) performing corner matching sorting, performing Euclidean distance calculation by utilizing the orderly coordinates of the building outline points and the corner coordinates, taking the Euclidean distance calculation as similarity measurement of corner matching, and taking the serial number of the point on the building outline with the minimum Euclidean distance as the serial number of the matched corner;
(3) the angular point characteristic analysis and elimination are carried out, because a point set obtained by angular point detection can generate fine line segments and redundant angular points with high precision, in addition, the angular points which are shielded by small shadows or are wrongly extracted and do not accord with morphological rules during building extraction also need to be eliminated, two-line included angles theta formed by three adjacent angular points are sequentially and iteratively calculated according to the angular point sequence numbers, and an included angle set T is set as { alpha-80 DEG<α<80 degrees, when the included angle theta belongs to T, approximately considering that P1 is a non-corner point, and deleting P1 if three points P0, P1 and P2 possibly lie on the same line segment; when the angle is included
Reserving a point P1; gradually iterating and calculating according to the method, eliminating unnecessary angular points and remaining the unnecessary angular pointsThe angular points are orderly connected, and finally, the comprehensive arrangement of the building is achieved;
according to the technical scheme, hierarchical optimization is carried out, firstly, building outline primary extraction is carried out by utilizing a complete convolution neural network, and then on the basis, the building initial outline is extracted to carry out polygon primary fitting processing; secondly, performing primary regularization on the fitted contour by using a best fit external rectangular junction Hausdorff distance algorithm; and finally, performing depth optimization on the local contour which cannot be optimized in the complex contour by using a Shi-Tomasi algorithm. And finally outputting the optimal and orderly building contour result.
Compared with the prior art, the invention has the beneficial effects that: the invention has scientific and reasonable structure and safe and convenient use, extracts the building by utilizing the characteristics of the building such as spectral information, shape texture and the like, performs multi-stage optimization on the complex building outline by full convolution neural network, polygonal fitting, best fitting external rectangle, Hausdorff distance algorithm and Shi-Tomasi algorithm, can effectively improve the extraction accuracy, thereby fully utilizing the network structure of the space context information under different scales, optimizing the dynamic range of the building size, reducing the network parameters, improving the image processing efficiency and finally improving the accuracy of the building outline extraction by optimizing the full convolution neural network algorithm.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. A building surface multistage optimization extraction method based on a full convolution neural network is characterized by comprising the following steps: the method comprises the following steps:
s1, preliminarily extracting the outline of the building by using a full convolution-based neural network, and then extracting the initial outline of the building on the basis of the preliminary outline of the building to perform polygon fitting treatment;
s2, performing initial contour warping on the fitted contour by utilizing a best fit external rectangular junction Hausdorff distance algorithm;
and S3, finally, carrying out depth optimization on the local contour which cannot be optimized in the complex contour by using a Shi-Tomasi algorithm, and finally outputting the best optimization regular building contour result.
2. The method for multi-level optimized extraction of building surfaces based on full convolution neural network as claimed in claim 1, wherein the specific steps in S1 are removing all FC layers of FCNs network;
and changing feature maps with different sizes in level2 into the size of an original image by using different up-sampling magnifications, and then obtaining a prediction result through a full convolution layer.
3. The method for multistage optimization extraction of building surfaces based on a full convolution neural network as claimed in claim 1, wherein in S1, building contour extraction is performed, and polygon fitting is performed on the contour by using a Douglas-Peucker algorithm, and the polygon fitting by the Douglas-Peucker algorithm, because there are saw teeth on the edge of the initial building contour extracted by the classification principle, the number of contour points is large, which is not favorable for calculation of the subsequent contour points, and in order to reduce the calculation amount, the main shape features of the contour are retained by the polygon fitting method.
4. The building surface multistage optimization extraction method based on the full convolution neural network as claimed in claim 1, wherein the best-fit circumscribed rectangle in S2 is extracted, and the circumscribed rectangle consistent with the inclination degree of the building axis is used as the best-fit circumscribed rectangle, so that the accuracy can be effectively improved.
5. The method for multi-stage optimization extraction of building surfaces based on full convolution neural network as claimed in claim 1, wherein said S2 utilizes Hausdorff distance algorithm to accurately measure the distance difference between the fitting contour of the building and the best fitting circumscribed rectangle as the criterion for judging whether the fitting contour is suitable.
6. The method for multistage optimization extraction of building surfaces based on a full convolution neural network as claimed in claim 1, wherein in S3, the contour of a complex building is optimized based on a Shi-Tomasi algorithm, and for right-angle corner characteristics of a right-angle building, corner detection is used and corner selection and deletion are performed, so that sharp small angles can be removed while the right-angle details of the contour are retained, and thereby vertical corners conforming to the contour of the building are obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011424539.1A CN112487537A (en) | 2020-12-08 | 2020-12-08 | Building surface multistage optimization extraction method based on full convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011424539.1A CN112487537A (en) | 2020-12-08 | 2020-12-08 | Building surface multistage optimization extraction method based on full convolution neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112487537A true CN112487537A (en) | 2021-03-12 |
Family
ID=74940884
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011424539.1A Pending CN112487537A (en) | 2020-12-08 | 2020-12-08 | Building surface multistage optimization extraction method based on full convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112487537A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113744144A (en) * | 2021-08-20 | 2021-12-03 | 长江大学 | Remote sensing image building boundary optimization method, system, equipment and storage medium |
| CN114387329A (en) * | 2022-01-16 | 2022-04-22 | 四川轻化工大学 | Building contour progressive regularization method based on high-resolution remote sensing image |
| CN115019163A (en) * | 2022-04-01 | 2022-09-06 | 湖南大学 | City factor identification method based on multi-source big data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070195344A1 (en) * | 2006-02-01 | 2007-08-23 | Sony Corporation | System, apparatus, method, program and recording medium for processing image |
| US20080056572A1 (en) * | 2006-08-29 | 2008-03-06 | Frank Nielsen | Image processing apparatus, image processing method and computer program |
| CN101419669A (en) * | 2008-10-14 | 2009-04-29 | 复旦大学 | Three-dimensional human ear extracting method based on profile wave convert |
| CN102938066A (en) * | 2012-12-07 | 2013-02-20 | 南京大学 | Method for reconstructing outer outline polygon of building based on multivariate data |
| US20200027266A1 (en) * | 2018-07-17 | 2020-01-23 | Uti Limited Partnership | Building contour generation from point clouds |
-
2020
- 2020-12-08 CN CN202011424539.1A patent/CN112487537A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070195344A1 (en) * | 2006-02-01 | 2007-08-23 | Sony Corporation | System, apparatus, method, program and recording medium for processing image |
| US20080056572A1 (en) * | 2006-08-29 | 2008-03-06 | Frank Nielsen | Image processing apparatus, image processing method and computer program |
| CN101419669A (en) * | 2008-10-14 | 2009-04-29 | 复旦大学 | Three-dimensional human ear extracting method based on profile wave convert |
| CN102938066A (en) * | 2012-12-07 | 2013-02-20 | 南京大学 | Method for reconstructing outer outline polygon of building based on multivariate data |
| US20200027266A1 (en) * | 2018-07-17 | 2020-01-23 | Uti Limited Partnership | Building contour generation from point clouds |
Non-Patent Citations (2)
| Title |
|---|
| 常京新等: "高分遥感影像建筑物轮廓的逐级优化方法", 《中国激光》, vol. 47, no. 10, pages 249 - 262 * |
| 王双喜;杨元维;常京新;高贤君;: "高分辨率影像分类提取建筑物轮廓的优化方法", 激光与光电子学进展, no. 02, pages 329 - 338 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113744144A (en) * | 2021-08-20 | 2021-12-03 | 长江大学 | Remote sensing image building boundary optimization method, system, equipment and storage medium |
| CN113744144B (en) * | 2021-08-20 | 2023-09-26 | 长江大学 | Remote sensing image building boundary optimization method, system, equipment and storage medium |
| CN114387329A (en) * | 2022-01-16 | 2022-04-22 | 四川轻化工大学 | Building contour progressive regularization method based on high-resolution remote sensing image |
| CN114387329B (en) * | 2022-01-16 | 2024-03-22 | 四川轻化工大学 | Building contour progressive regularization method based on high-resolution remote sensing image |
| CN115019163A (en) * | 2022-04-01 | 2022-09-06 | 湖南大学 | City factor identification method based on multi-source big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112487537A (en) | Building surface multistage optimization extraction method based on full convolution neural network | |
| CN113450307B (en) | Product edge defect detection method | |
| CN115082419B (en) | Blow-molded luggage production defect detection method | |
| CN106022381B (en) | Automatic extraction method of street lamp pole based on vehicle-mounted laser scanning point cloud | |
| CN109377485B (en) | Machine vision detection method for instant noodle packaging defects | |
| CN111626146B (en) | Merging cell table segmentation recognition method based on template matching | |
| CN113781402A (en) | Method and device for detecting chip surface scratch defects and computer equipment | |
| CN116596936B (en) | Solar wing sailboard quality detection method based on image data | |
| CN110188778B (en) | Residential area element outline regularization method based on image extraction result | |
| US11037299B2 (en) | Region merging image segmentation algorithm based on boundary extraction | |
| CN112801075B (en) | Automatic rural road boundary line extraction method based on aerial image | |
| CN111091071B (en) | Underground target detection method and system based on ground penetrating radar hyperbolic wave fitting | |
| CN112686265A (en) | Hierarchic contour extraction-based pictograph segmentation method | |
| CN114898118A (en) | Automatic statistical method and system for power transmission line house removal amount based on multi-source point cloud | |
| CN111275732B (en) | Foreground object image segmentation method based on depth convolution neural network | |
| CN109829511B (en) | Texture classification-based method for detecting cloud layer area in downward-looking infrared image | |
| CN109117841B (en) | Scene text detection method based on stroke width transformation and convolutional neural network | |
| CN110751150A (en) | FPGA-based binary neural network license plate recognition method and system | |
| CN115830587A (en) | Structural surface rapid automatic identification method based on high-precision point cloud data | |
| CN115063698A (en) | Automatic identification and information extraction method and system for slope surface deformation crack | |
| CN110532826B (en) | Bar code recognition device and method based on artificial intelligence semantic segmentation | |
| CN114120140A (en) | Method for automatically extracting building height based on satellite image | |
| CN110276260B (en) | Commodity detection method based on depth camera | |
| CN111192264A (en) | Automatic paper defect detection method based on improved non-local mean value fast-RCNN algorithm | |
| CN112183229A (en) | Character lattice extraction method and device of job paper image based on calculation of dynamic parameters |
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 |