CN109446992B - Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment - Google Patents
Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN109446992B CN109446992B CN201811279197.1A CN201811279197A CN109446992B CN 109446992 B CN109446992 B CN 109446992B CN 201811279197 A CN201811279197 A CN 201811279197A CN 109446992 B CN109446992 B CN 109446992B
- Authority
- CN
- China
- Prior art keywords
- building
- remote sensing
- image
- sensing image
- data
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention provides a remote sensing image building extraction method based on deep learning, which comprises the steps of sample preparation, model training, precision evaluation, building prediction, merging and vectorization. The invention also relates to a remote sensing image building extraction system, a storage medium and electronic equipment based on deep learning; the method comprises the steps of extracting urban monomer building outlines, rural isolated building outlines and rural building dense cluster peripheral boundaries on the basis of an improved RCF boundary constraint model, improving a U-Net semantic segmentation model network structure, performing pixel-level classification on images by using the improved U-Net, finally fusing the two models, training a deep learning model through a large amount of building sample label data, and extracting buildings on sub-meter domestic high-grade No. 2 remote sensing images by using the improved U-Net and RCF fused network model, so that automatic and efficient building vector data extraction is realized, and the time cost and the labor cost of manual drawing are greatly reduced.
Description
Technical Field
The invention belongs to the technical field of remote sensing image processing, and relates to a method for extracting a high-spatial-resolution remote sensing image building based on a deep learning boundary constraint algorithm, which is mainly applied to automatic building extraction of a sub-meter-level remote sensing image.
Background
Buildings are the most easily added and changed parts of the geographic database and the most in need of updating. Due to the importance of the building to the aspects of city construction, GIS system updating, digital cities, military reconnaissance and the like, the technology for quickly extracting the building information and the detection of the building change have important application in the aspects of city development planning, electronic informatization, national defense and the like. The extraction of the information of the artificial building in the remote sensing image is a complex process, not only needs the automatic identification of a computer, but also needs the assistance of people to complete, and therefore the efficiency of the extraction of the building at present is low.
In recent years, with the great improvement of computer performance and the rapid development of deep learning, the application field of the convolutional neural network is continuously expanded, the method for automatically extracting the ground features by using the convolutional neural network is gradually mature, and great effect is achieved in the field of remote sensing. Due to the improvement of the accuracy of the sensor, the resolution ratio of the usable remote sensing image is higher and higher, and the features extracted by the convolutional neural network are richer and richer. Therefore, the method has theoretical foundation for extracting various buildings on the high-resolution remote sensing image by using the convolutional neural network and taking the buildings as units.
The existing technical method for extracting buildings based on remote sensing images only can extract regular buildings with obvious characteristics, has poor universality, and has a common extraction effect when the buildings are dense, so that how to quickly and accurately extract various complex buildings is a key step for processing remote sensing information.
At present, the difference between a building vector obtained by adopting a deep learning building extraction method and an actual building block represented by an image is large, and the method is difficult to apply to land utilization investigation. At present, a novel remote sensing image building extraction method is urgently needed.
Disclosure of Invention
In order to overcome the defects of the prior art, the remote sensing image building extraction method based on deep learning provided by the invention adopts the improved network model fusing the U-Net and the RCF, so that buildings of various types can be quickly detected on the remote sensing image, the extracted building boundary can be effectively constrained, and the consistency of the extracted building boundary and the actual image building boundary is ensured.
The invention provides a remote sensing image building extraction method based on deep learning, which comprises the following steps:
s1, sample preparation, namely collecting a remote sensing image, cutting the remote sensing image to obtain a surface vector file containing a classified object, drawing a building label in the surface vector file, and converting vector data of the surface vector sample labeled as a building into raster data to obtain a rasterized building sample, wherein the building sample comprises a city single building, a rural isolated building and a rural homestead dense building group;
s2, model training, namely adjusting improved algorithm model parameters after combination of RCF and U-Net, and performing improved algorithm model training on the building sample based on combination of RCF and U-Net to obtain three classification models corresponding to boundary constraints of the urban single building, the rural isolated building and the rural homestead dense building group;
s3, precision evaluation, namely inputting remote sensing image data to be tested into the classification model prediction test data, calculating a detection evaluation function MIoU and MPA of the test, jumping to the next step if the MIoU value or the MPA both reach the standard, adjusting the improved U-Net and RCF fused network model parameters to return to the step S1 and modifying the sample for iterative training if the MIoU value or the MPA does not reach the standard;
s4, building prediction, namely, performing building prediction on the remote sensing image target area by using the three classification models which are evaluated to reach the standard in the step S3 to obtain three types of gridded building data in the remote sensing image;
and S7, merging and vectorizing, namely superposing the three final rasterized building data to obtain total building raster data, and vectorizing the total building raster data to obtain the extracted remote sensing image building data.
Further, between the step S4 and the step S7, the method further includes the steps of:
and S5, performing image post-processing, namely performing boundary smoothing on the three types of rasterized building data predicted and obtained in the step S4 by using an image processing module, and reducing building boundary jaggies.
Further, between the step S4 and the step S7, the method further includes the steps of:
s6, building pattern spot processing, namely acquiring a pattern spot judgment threshold value according to the building type, judging that the pattern spot is an effective pattern spot if the area of the pattern spot corresponding to the building type is larger than or equal to the pattern spot judgment threshold value, and repairing the pattern spot by performing pattern spot corrosion, opening and closing operation and filling processing; and if the area of the image spots corresponding to the building types is smaller than the image spot judgment threshold value, judging the image spots as invalid image spots, and carrying out image spot corrosion and denoising treatment to eliminate the filtered image spots.
Further, the building label includes a line label and a surface label.
Further, in step S1, the boundary of the remote sensing image is extracted through the RCF algorithm, building classification is performed according to the boundary, and the boundary of the remote sensing image extracted through the RCF algorithm is used as a constraint condition and is input into the building sample.
Further, the remote sensing image is a three-band or four-band fusion image.
An electronic device, comprising: a processor;
a memory; and a program, wherein the program is stored in the memory and configured to be executed by the processor, the program comprising instructions for performing a remote sensing imagery building extraction method based on deep learning.
A computer-readable storage medium having stored thereon a computer program for execution by a processor of a method for remote sensing image building extraction based on deep learning.
The remote sensing image building extraction system based on deep learning comprises a sample making module, a model training module, a precision evaluation module, a building prediction module and an image processing module; wherein the content of the first and second substances,
the image processing module comprises a merging vectorization unit and a rasterization unit, wherein the merging vectorization unit is used for merging the rasterized data of the remote sensing image and then carrying out vectorization, and the rasterization unit is used for rasterizing the vectorized data of the remote sensing image;
the sample making module is used for acquiring a remote sensing image, cutting the remote sensing image to obtain a surface vector sample containing a classified object, drawing a label of the surface vector sample, converting vector data of the surface vector sample, which is labeled as a building, into raster data to obtain a rasterized building sample, wherein the building sample comprises a city single building, a rural isolated building and a rural homestead dense building group;
the model training module is used for adjusting improved algorithm model parameters after the combination of RCF and U-Net, and carrying out improved algorithm model training on the building sample based on the combination of RCF and U-Net to obtain three classification models corresponding to boundary constraints of city single buildings, rural isolated buildings and rural homestead dense building groups;
the precision evaluation module is used for inputting remote sensing image data to be tested into classification model prediction test data, calculating a detection evaluation function MIoU and an MPA of the test, jumping to the next step if the MIoU value or the MPA both reach the standard, and adjusting improved U-Net and RCF fused network model parameters to return and modifying a sample for iterative training if the MIoU value or the MPA does not reach the standard;
the building prediction module is used for carrying out building prediction on a remote sensing image target area by utilizing the classification model which is evaluated to reach the standard in the precision evaluation module to obtain three types of gridded building data in the remote sensing image;
and the merging vectorization unit superposes the three final rasterized building data to obtain total building raster data, and then vectorizes the total building raster data to obtain the extracted remote sensing image building data.
Further, the remote sensing image building extraction system based on deep learning further comprises a building pattern recognition module, wherein the building pattern recognition module is used for acquiring a pattern judgment threshold value according to the building type, and if the pattern area corresponding to the building type is larger than or equal to the pattern judgment threshold value, the pattern is judged to be an effective pattern; if the area of the image spot corresponding to the building type is smaller than the image spot judgment threshold value, judging the image spot to be an invalid image spot;
the image processing module also comprises an image post-processing unit, an invalid pattern spot processing unit and an effective pattern spot processing unit; the image post-processing unit is used for performing boundary smoothing on the three types of rasterized building data obtained by prediction in the building prediction module, and reducing building boundary sawteeth; the invalid pattern spot processing unit is used for carrying out pattern spot corrosion and denoising processing on the determined invalid pattern spots to eliminate the filtered pattern spots; the effective pattern spot processing unit is used for carrying out pattern spot corrosion, opening and closing operation and filling processing on the determined effective pattern spots to repair the pattern spots.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a remote sensing image building extraction method based on deep learning, which comprises the steps of sample preparation, model training, precision evaluation, building prediction, merging and vectorization. The invention also relates to a remote sensing image building extraction system, a storage medium and electronic equipment based on deep learning; the method adopts a deep optimization full convolution neural network, optimizes the network by improving parameters such as network layers, pooling size and the like of the model, and simultaneously adds a deep edge model to improve the extraction effect of the building outline. The method comprises the steps of extracting city monomer building outlines, rural isolated building outlines and rural building dense cluster peripheral boundaries on the basis of an improved RCF boundary constraint model, improving a U-Net semantic segmentation model network structure, utilizing the improved U-Net to classify images in a pixel level mode, finally fusing the two models, training a deep learning model through a large amount of building sample label data, extracting buildings on sub-meter domestic high-resolution No. 2 remote sensing images through an improved algorithm model after the RCF and the U-Net are combined, achieving automatic and efficient building vector data extraction, and greatly reducing time cost and labor cost of manual drawing. Compared with the traditional artificial building block extraction method, the method can effectively improve the production efficiency of building block extraction, simultaneously ensures the consistency of the form of the building monomer and the form of the building displayed on the remote sensing image, and is convenient for remote sensing image processing, popularization and application.
The type and the form of the building on the remote sensing image are complex, the color and the shape are different, the scale difference is large, a single house with a small rural homebase is provided, and a factory building with a large floor area is also provided. Aiming at the complex characteristics of the building, a U-Net and RCF fusion network model is utilized, all the convolution layer information in the network is utilized, and multi-scale fusion is explicitly used to fully extract object characteristics including the external contour characteristics of the building and the internal texture characteristics of the building.
The foregoing description is only an overview of the technical solutions of the present invention, and in order to make the technical solutions of the present invention more clearly understood and to implement them in accordance with the contents of the description, the following detailed description is given with reference to the preferred embodiments of the present invention and the accompanying drawings. The detailed description of the present invention is given in detail by the following examples and the accompanying drawings.
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 application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
FIG. 1 is a schematic flow chart of a remote sensing image building extraction method based on deep learning according to the present invention;
FIG. 2 is a logic diagram of the remote sensing image building extraction method based on deep learning according to the present invention;
FIG. 3 is a schematic diagram of a deep learning algorithm U-Net model;
FIG. 4 is a schematic diagram of an RCF algorithm model;
FIG. 5 is a schematic diagram of an improved U-Net and RCF fusion model of the present invention;
FIG. 6 is a partial schematic view of the building of FIG. 5 after the present invention has been applied;
fig. 7 is a schematic diagram of the remote sensing image building extraction system based on deep learning of the present invention.
Detailed Description
The present invention will be further described with reference to the accompanying drawings and the detailed description, and it should be noted that any combination of the embodiments or technical features described below can be used to form a new embodiment without conflict.
The remote sensing image building extraction method based on deep learning is shown in fig. 1 and 2 and comprises the following steps:
s1, sample preparation, namely collecting a remote sensing image, cutting the remote sensing image to obtain a surface vector file containing a classified object, drawing a building label in the surface vector file, and converting vector data of the surface vector sample labeled as a building into raster data to obtain a rasterized building sample, wherein the building sample comprises a city single building, a rural isolated building and a rural homestead dense building group.
In one embodiment, a high-resolution No. 2 0.8-meter resolution 4-waveband fusion image of a research area is collected and sorted, an image library is established, and meanwhile, a corresponding building vector label sample library (comprising urban single buildings, rural isolated buildings and rural homestead dense building groups) is established according to the image; and carrying out format conversion of the sample label according to the building sample library, and converting the building sample into contour line sample raster data and building surface sample raster data according to the label vector. For example, automatically cutting an image sample block on a high-score second-grade remote sensing image through a selected point vector file to obtain a surface vector frame sample which is 1000 × 1000 in size and contains a classification object, drawing a sample label, and processing multiband sample Data in batch by using GDAL (geographic Data Abstraction library) into a standard format, wherein the sample types comprise urban single buildings, rural isolated buildings and rural homestead dense building groups, and each sample comprises two types of labels of lines and surfaces; it should be noted that, in the present embodiment, a four-band fusion image is preferred, and since the four-band fusion image further includes a near-infrared image on the basis of an RGB three-band image, the data amount of the data sample is larger than that of a common three-band image, which is beneficial to sample training, thereby improving accuracy; secondly, because the near-infrared image can effectively filter out green interference, and the image information of the green belt can be effectively and quickly filtered after being combined with the RGB three wave bands, the building extraction efficiency is improved. In the embodiment, the boundary of the remote sensing image is extracted by the contour line of the building through an RCF (Richer connected features) algorithm, the remote sensing image is subjected to pixel-level classification through improved U-Net by building face sample raster data, and the boundary is used as a constraint and input into a deep learning network model.
And S2, training a model, adjusting improved algorithm model parameters after the combination of RCF and U-Net, and performing improved algorithm model training on the building sample based on the combination of RCF and U-Net to obtain three classification models corresponding to the boundary constraints of the urban single building, the rural isolated building and the rural homestead dense building group.
As shown in FIG. 3, for the deep learning algorithm U-Net model principle, U-Net is a multi-scale network model obtained by adding an up-sampling stage on the basis of an FCN (fuzzy C-means) network, and is suitable for a remote sensing large-figure semantic segmentation convolutional neural network, the network structure is mainly divided into two parts as shown in FIG. 3, an input side part is a convolutional network for extracting image features like the FCN, an output side part is the deconvolution operation of the up-sampling part, and a plurality of feature channels are added in the stage, so that more original image texture information is allowed to be transmitted in a high-resolution layer. Particularly, U-Net has no fully connected layer, and the whole process uses valid to perform convolution, so that the segmentation result is guaranteed to be obtained based on the context features without missing, and therefore, the image sizes of input and output are inconsistent, and for the input of a large image, the overlap-protocol can be used to perform seamless image output. In addition, the upsampling part fuses the output of the feature extraction part, so that the multi-scale fusion is realized, the features of the object are both the output from the first convolution block and the upsampled output, and the connection runs through the whole network and has a good extraction effect on buildings with various scales.
As shown in fig. 4, for the RCF algorithm model principle, the RCF algorithm is a certain improvement on the basis of the HED algorithm structure, wherein the HED algorithm adopts a cascade structure and simultaneously uses the VGG as a model to perform fine-tune, outputs the outputs of 5 stages in the VGG for fusion, but the HED outputs only the convolution characteristic of the last layer in each stage, while the RCF outputs the characteristics of all convolution layers in each stage and introduces a new loss function loss/sigmoid.
FIG. 5 shows an improved algorithm model principle after combining a deep learning-based boundary constraint model, namely RCF and U-Net; the RCF algorithm principle penetrates through the whole U-Net network, 1 x 1 convolution kernel is used behind each convolution layer of the U-Net network for dimension increasing or dimension reducing and information integration among channels, the feature layers in each stage are fused, loss is calculated for each fused layer, all edge loss is weighted and averaged, finally, the weighted average is carried out on all edge loss and the classified loss to obtain final loss, the final output result is the result of fusion of the last convolution output of an edge module and the last convolution output of a classification module, and boundary features are fully utilized for constraint.
In an embodiment, model training based on boundary constraint of deep learning is performed according to the three converted classification samples, the pooling size of the number of network layers in the model is adjusted, and feature fusion is performed. For example, the image grid data of each building and the corresponding two types of labels are simultaneously input into the improved network of U-Net and RCF shown in FIG. 5, the characteristics of the model to the boundary and the internal characteristics of the building are fully considered, the boundary constraint is realized while the classification is carried out, and the classification models of the boundary constraint corresponding to the three types of buildings are respectively obtained by carrying out repeated iterative training through adjusting parameters.
S3, precision evaluation, namely inputting remote sensing image data to be tested into the classification model prediction test data, calculating a detection evaluation function MIoU and MPA of the test, jumping to the next step if the MIoU value or the MPA both reach the standard, adjusting the improved U-Net and RCF fused network model parameters to return to the step S1 and modifying the sample for iterative training if the MIoU value or the MPA does not reach the standard; for example, if the MIoU value is set to be in the range of 0.80-1, the MPA value is set to be in the range of 0.85-1, the network model is judged to be in the standard, and if the MIoU value is out of the range, the improved U-Net and RCF fused network model parameters are adjusted and returned to S1 for retraining.
S4, building prediction, namely, performing building prediction on the remote sensing image target area by using the three classification models which are evaluated to reach the standard in the step S3 to obtain three types of gridded building data in the remote sensing image; for example, the remote sensing image target area high-grade No. 2 remote sensing image is subjected to building extraction by using an improved algorithm model after RCF and U-Net are combined, the characteristics of the model to the boundary and the internal characteristics of the building are fully considered, and boundary constraint is realized while classification is carried out.
S7, merging and vectorizing, namely superposing the three final rasterized building data to obtain total building raster data, vectorizing the total building raster data to obtain extracted remote sensing image building data so as to obtain an applicable result, and finally carrying out vectorization operations such as topology inspection, edge inspection and the like on the vectorized remote sensing image building data; for example, fig. 6 is a schematic diagram illustrating the result of extracting the buildings in the urban part of the canxia region in Nanjing, wherein the black outlines are all the boundaries of the buildings. Through the verification of the actual artificial building production test, the technical method of the invention is greatly improved in efficiency: the area of a test area sample is about 200 square kilometers, 16231 buildings (groups) are used in total, about 48 hours are consumed for single extraction, 1 hour is consumed for extraction by applying the technical method of the invention, 1.5 hours is consumed for manual fine trimming, and the production efficiency is improved by about 20 times. Therefore, the invention can effectively improve the efficiency of the existing building extraction method on the premise of ensuring the precision.
It should be noted that the vector data structure can be specifically divided into points, lines, and planes, and can constitute various complex entities in the real world, and is particularly effective when the problem can be described as a line or a boundary. The vector data has compact structure and low redundancy, has the topological information of the space entity, is easy to define and operate a single space entity, and is convenient for network analysis. The output quality of the vector data is good and the precision is high. However, the complexity of the vector data structure leads to complexity of operation and algorithm, and as a coding method based on lines and boundaries, the vector data structure cannot effectively support image algebraic operation, for example, cannot effectively perform collective operation (such as superposition) of point sets, and has low and complicated operation efficiency. The storage of the vector data structure is complex, so that the query of the space entity is time-consuming and needs to be queried point by point, line by line and plane by plane. Vector data and raster-represented image data cannot be directly operated (such as joint query and spatial analysis), and vector and raster conversion must be carried out during interaction. The interaction of the vector data with the DEM (digital elevation model) is realized by contour lines and cannot be directly subjected to joint spatial analysis with the DEM. The raster data structure represents the overall spatial phenomenon by a dense and regular arrangement of spatial points. The data structure is simple, the positioning access performance is good, the combined spatial analysis can be carried out on the image and DEM data, the data sharing is easy to realize, and the operation on raster data is easier. The amount of raster data is inversely proportional to the square of the grid pitch, and the higher geometric accuracy comes at the cost of a greatly increased amount of data. Because only rows and columns are used as the position identifiers of the space entities, it is difficult to obtain topology information of the space entities, and it is difficult to perform operations such as network analysis. The raster data structure is not entity-oriented, and various entities are often reflected by being overlaid together, so that the entities are difficult to identify and separate. The identification of the point entity needs to adopt a matching technology, the identification of the line entity needs to adopt an edge detection technology, and the identification of the opposite entity needs to adopt an image classification technology, which not only wastes time, but also can not ensure complete correctness. It can be seen from the above analysis that the advantages and disadvantages of the vector data structure and the raster data structure are complementary, and in order to effectively implement each function in the GIS (such as combination with remote sensing image data, effective spatial analysis, etc.), two data structures need to be used simultaneously, and efficient conversion of the two data structures in the GIS is implemented.
In an embodiment, between the step S4 and the step S7, the method further includes the steps of:
and S5, performing image post-processing, namely performing boundary smoothing on the three types of rasterized building data predicted and obtained in the step S4 by using an image processing module, and reducing building boundary jaggies. In this embodiment, the extracted building contour boundary is first smoothed and refined. It should be understood that the image processing includes, but is not limited to, existing graphical algorithms such as edge smoothing and thinning processing, and in the present embodiment, the prediction result is post-processed by using an opencv image processing module.
In an embodiment, an originally coherent building outage caused by a deviation easily occurred in the building extraction process, the method further includes, between the step S4 and the step S7:
s6, building pattern spot processing, namely acquiring a pattern spot judgment threshold value according to the building type, judging that the pattern spot is an effective pattern spot if the area of the pattern spot corresponding to the building type is larger than or equal to the pattern spot judgment threshold value, and repairing the pattern spot by performing pattern spot corrosion, opening and closing operation and filling processing; and if the area of the image spots corresponding to the building types is smaller than the image spot judgment threshold value, judging the image spots as invalid image spots, and carrying out image spot corrosion and denoising treatment to eliminate the filtered image spots. In this embodiment, the extracted size of the architectural pattern spot is judged first, the architectural single body and the architectural group are distinguished, and the hole repairing of the architectural group and the denoising of the architectural single body are realized through a graphical algorithm; for example, the image spots are divided into effective image spots and ineffective image spots, the judgment standard is an empirical value which is mainly divided according to the area size of buildings, the average sizes of the three buildings are different, so that the image spot judgment thresholds of different types of buildings are also different, and when the effective image spots are judged, a series of image operations including but not limited to image spot corrosion, opening and closing operation and filling are carried out; and when the pattern spots are judged to be invalid, eliminating the pattern spots by a corrosion and denoising method, and filtering a part of error extraction results.
An electronic device, comprising: a processor; a memory; and a program, wherein the program is stored in the memory and configured to be executed by the processor, the program comprising instructions for performing a remote sensing imagery building lift method based on deep learning. A computer-readable storage medium having stored thereon a computer program for execution by a processor of a method for remote sensing image building extraction based on deep learning.
The remote sensing image building extraction system based on deep learning is shown in fig. 7 and comprises a sample making module, a model training module, a precision evaluation module, a building prediction module and an image processing module; wherein the content of the first and second substances,
the image processing module comprises a merging vectorization unit and a rasterization unit, wherein the merging vectorization unit is used for merging the rasterized data of the remote sensing image and then carrying out vectorization, and the rasterization unit is used for rasterizing the vectorized data of the remote sensing image;
the system comprises a sample making module, a data processing module and a data processing module, wherein the sample making module is used for acquiring a remote sensing image, segmenting and cutting the remote sensing image to obtain a surface vector sample containing a classified object, drawing a label of the surface vector sample, converting vector data of the surface vector sample, which is labeled as a building, into raster data to obtain a rasterized building sample, and the type of the building sample comprises an urban single building, a rural isolated building and a rural homestead dense building group;
the model training module is used for adjusting improved algorithm model parameters after the RCF and the U-Net are combined, and performing improved algorithm model training on a building sample based on the RCF and the U-Net to obtain three classification models corresponding to boundary constraints of city single buildings, rural isolated buildings and rural homeland dense building groups;
the precision evaluation module is used for inputting remote sensing image data to be tested into classification model prediction test data, calculating a detection evaluation function MIoU and an MPA of the test, jumping to the next step if the MIoU value or the MPA both reach the standard, and adjusting improved U-Net and RCF fused network model parameters to return and modifying a sample for iterative training if the MIoU value or the MPA does not reach the standard;
the building prediction module is used for carrying out building prediction on a remote sensing image target area by utilizing the classification model which is evaluated to reach the standard in the precision evaluation module to obtain three types of gridded building data in the remote sensing image;
and the merging vectorization unit superposes the three final rasterized building data to obtain total building raster data, and then vectorizes the total building raster data to obtain the extracted remote sensing image building data.
Preferably, as shown in fig. 7, the remote sensing image building extraction system based on deep learning further includes a building pattern recognition module, where the building pattern recognition module is configured to obtain a pattern determination threshold according to a building category, and if a pattern area corresponding to the building category is greater than or equal to the pattern determination threshold, determine that a pattern is an effective pattern; if the area of the image spot corresponding to the building type is smaller than the image spot judgment threshold value, judging the image spot to be an invalid image spot;
the image processing module also comprises an image post-processing unit, an invalid pattern spot processing unit and an effective pattern spot processing unit; the image post-processing unit is used for performing boundary smoothing on the three types of rasterized building data obtained by prediction in the building prediction module, and reducing building boundary sawteeth; the invalid pattern spot processing unit is used for carrying out pattern spot corrosion and denoising processing on the determined invalid pattern spots to eliminate the filtered pattern spots; the effective pattern spot processing unit is used for carrying out pattern spot corrosion, opening and closing operation and filling processing on the determined effective pattern spots to repair the pattern spots.
The invention provides a remote sensing image building extraction method based on deep learning, which adopts a deep optimization full convolution neural network, optimizes the network by improving parameters such as network layers, pooling size and the like of a model, and simultaneously adds a deep edge model to improve the building outline extraction effect. The method comprises the steps of extracting urban monomer building outlines, rural isolated building outlines and rural building dense cluster peripheral boundaries on the basis of an improved RCF boundary constraint model, improving a U-Net semantic segmentation model network structure, performing pixel-level classification on images by utilizing the improved U-Net, finally fusing the two models, training a deep learning model through a large amount of building sample label data, extracting buildings on sub-meter domestic high-grade No. 2 remote sensing images by using an improved algorithm model after the RCF and the U-Net are combined, realizing automatic and efficient building vector data extraction, and greatly reducing the time cost and the labor cost of manual drawing. Compared with the traditional artificial building block extraction method, the method can effectively improve the production efficiency of building block extraction, and simultaneously ensure the consistency of the form of the building monomer and the form of the building displayed on the remote sensing image.
The type and the form of the building on the remote sensing image are complex, the color and the shape are different, the scale difference is large, and the building has a small single-span house of a rural homestead and a factory building with a large floor area. Aiming at the complex characteristics of the building, a U-Net and RCF fusion network model is utilized, all the convolution layer information in the network is utilized, and multi-scale fusion is explicitly used to fully extract object characteristics including the external contour characteristics of the building and the internal texture characteristics of the building.
The foregoing is merely a preferred embodiment of the invention and is not intended to limit the invention in any manner; those skilled in the art can readily practice the invention as shown and described in the drawings and detailed description herein; however, those skilled in the art should appreciate that they can readily use the disclosed conception and specific embodiments as a basis for designing or modifying other structures for carrying out the same purposes of the present invention without departing from the scope of the invention as defined by the appended claims; meanwhile, any changes, modifications, and evolutions of the equivalent changes of the above embodiments according to the actual techniques of the present invention are still within the protection scope of the technical solution of the present invention.
Claims (10)
1. The remote sensing image building extraction method based on deep learning is characterized by comprising the following steps of:
s1, sample preparation, namely collecting a remote sensing image, cutting the remote sensing image to obtain a surface vector file containing a classified object, drawing a building label in the surface vector file, and converting vector data of the surface vector sample labeled as a building into raster data to obtain a rasterized building sample, wherein the building sample comprises a city single building, a rural isolated building and a rural homestead dense building group;
s2, model training, namely adjusting improved algorithm model parameters after combination of RCF and U-Net, and performing improved algorithm model training on the building sample based on combination of RCF and U-Net to obtain three classification models corresponding to boundary constraints of the urban single building, the rural isolated building and the rural homestead dense building group; the improved algorithm model after the RCF and the U-Net are combined is specifically as follows: the RCF algorithm principle penetrates through the whole U-Net network, 1 x 1 convolution kernel is used behind each convolution layer of the U-Net network for dimension increasing or dimension reducing and information integration among channels, the feature layers of each stage are fused, loss is calculated for each fused layer, all edge loss is weighted and averaged, finally, the weighted average is carried out on all edge loss and the classified loss to obtain final loss, the final output result is the result of fusing the final convolution output of the edge module and the final convolution output of the classification module, and boundary features are fully utilized for constraint;
s3, precision evaluation, namely inputting remote sensing image data to be tested into the classification model prediction test data, calculating a detection evaluation function MIoU and MPA of the test, jumping to the next step if the MIoU value or the MPA both reach the standard, adjusting the improved U-Net and RCF fused network model parameters to return to the step S1 and modifying the sample for iterative training if the MIoU value or the MPA does not reach the standard;
s4, building prediction, namely, performing building prediction on the remote sensing image target area by using the three classification models which are evaluated to reach the standard in the step S3 to obtain three types of gridded building data in the remote sensing image;
and S7, merging and vectorizing, namely superposing the three final rasterized building data to obtain total building raster data, and vectorizing the total building raster data to obtain the extracted remote sensing image building data.
2. The method for extracting remote sensing image buildings based on deep learning as claimed in claim 1, wherein the steps between S4 and S7 further comprise the steps of:
and S5, performing image post-processing, namely performing boundary smoothing on the three types of rasterized building data predicted and obtained in the step S4 by using an image processing module, and reducing building boundary jaggies.
3. The method for extracting remote sensing image buildings based on deep learning as claimed in claim 1 or 2, wherein between the step S4 and the step S7, the method further comprises the steps of:
s6, building pattern spot processing, namely acquiring a pattern spot judgment threshold value according to the building type, judging that the pattern spot is an effective pattern spot if the area of the pattern spot corresponding to the building type is larger than or equal to the pattern spot judgment threshold value, and repairing the pattern spot by performing pattern spot corrosion, opening and closing operation and filling processing; and if the area of the image spots corresponding to the building types is smaller than the image spot judgment threshold value, judging the image spots as invalid image spots, and carrying out image spot corrosion and denoising treatment to eliminate the filtered image spots.
4. The remote sensing image building extraction method based on deep learning of claim 3, characterized in that: the building label comprises a line label and a surface label.
5. The remote sensing image building extraction method based on deep learning of claim 3, characterized in that: in step S1, the boundary of the remote sensing image is extracted by the RCF algorithm, building classification is performed according to the boundary, and the boundary of the remote sensing image extracted by the RCF algorithm is used as a constraint condition and is input into the building sample.
6. The remote sensing image building extraction method based on deep learning of claim 4 or 5, characterized in that: the remote sensing image is a three-band or four-band fusion image.
7. An electronic device, characterized by comprising: a processor;
a memory; and a program, wherein the program is stored in the memory and configured to be executed by the processor, the program comprising instructions for carrying out the method of claim 1.
8. A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program is executed by a processor for performing the method as claimed in claim 1.
9. Remote sensing image building extraction system based on degree of depth learning, its characterized in that: the system comprises a sample making module, a model training module, an accuracy evaluation module, a building prediction module and an image processing module; wherein the content of the first and second substances,
the image processing module comprises a merging vectorization unit and a rasterization unit, wherein the merging vectorization unit is used for merging the rasterized data of the remote sensing image and then carrying out vectorization, and the rasterization unit is used for rasterizing the vectorized data of the remote sensing image;
the sample making module is used for acquiring a remote sensing image, cutting the remote sensing image to obtain a surface vector sample containing a classified object, drawing a label of the surface vector sample, converting vector data of the surface vector sample, which is labeled as a building, into raster data to obtain a rasterized building sample, wherein the building sample comprises a city single building, a rural isolated building and a rural homestead dense building group;
the model training module is used for adjusting improved algorithm model parameters after the combination of RCF and U-Net, and carrying out improved algorithm model training on the building sample based on the combination of RCF and U-Net to obtain three classification models corresponding to boundary constraints of city single buildings, rural isolated buildings and rural homestead dense building groups; the improved U-Net and RCF integrated network structure specifically comprises: the RCF algorithm principle penetrates through the whole U-Net network, 1 x 1 convolution kernel is used behind each convolution layer of the U-Net network for dimension increasing or dimension reducing and information integration among channels, the feature layers of each stage are fused, loss is calculated for each fused layer, all edge loss is weighted and averaged, finally, the weighted average is carried out on all edge loss and the classified loss to obtain final loss, the final output result is the result of fusing the final convolution output of the edge module and the final convolution output of the classification module, and boundary features are fully utilized for constraint;
the precision evaluation module is used for inputting remote sensing image data to be tested into classification model prediction test data, calculating a detection evaluation function MIoU and an MPA of the test, jumping to the next step if the MIoU value or the MPA both reach the standard, and adjusting improved network model parameters fused by U-Net and RCF to return and modifying a sample for iterative training again if the MIoU value or the MPA does not reach the standard;
the building prediction module is used for carrying out building prediction on a remote sensing image target area by utilizing the classification model which is evaluated to reach the standard in the precision evaluation module to obtain three types of gridded building data in the remote sensing image;
and the merging vectorization unit superposes the three final rasterized building data to obtain total building raster data, and then vectorizes the total building raster data to obtain the extracted remote sensing image building data.
10. The remote sensing image building extraction system based on deep learning of claim 9, characterized in that: the building pattern spot identification module is used for acquiring a pattern spot judgment threshold value according to the building type, and if the area of the pattern spot corresponding to the building type is larger than or equal to the pattern spot judgment threshold value, the pattern spot is judged to be an effective pattern spot; if the area of the image spot corresponding to the building type is smaller than the image spot judgment threshold value, judging the image spot to be an invalid image spot;
the image processing module also comprises an image post-processing unit, an invalid pattern spot processing unit and an effective pattern spot processing unit; the image post-processing unit is used for performing boundary smoothing on the rasterized three types of building data obtained by prediction in the building prediction module, and reducing building boundary sawteeth; the invalid image spot processing unit is used for carrying out image spot corrosion and denoising processing on the image spots judged as invalid to eliminate the filtering image spots; the effective pattern spot processing unit is used for carrying out pattern spot corrosion, opening and closing operation and filling processing on the determined effective pattern spots to repair the pattern spots.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811279197.1A CN109446992B (en) | 2018-10-30 | 2018-10-30 | Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811279197.1A CN109446992B (en) | 2018-10-30 | 2018-10-30 | Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109446992A CN109446992A (en) | 2019-03-08 |
| CN109446992B true CN109446992B (en) | 2022-06-17 |
Family
ID=65549932
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811279197.1A Active CN109446992B (en) | 2018-10-30 | 2018-10-30 | Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109446992B (en) |
Families Citing this family (48)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175548B (en) * | 2019-05-20 | 2022-08-23 | 中国科学院光电技术研究所 | Remote sensing image building extraction method based on attention mechanism and channel information |
| CN110197147B (en) * | 2019-05-23 | 2022-12-02 | 星际空间（天津）科技发展有限公司 | Building example extraction method, device, storage medium and equipment of remote sensing image |
| CN110298348B (en) * | 2019-06-12 | 2020-04-28 | 苏州中科天启遥感科技有限公司 | Method and system for extracting remote sensing image building sample region, storage medium and equipment |
| CN110580443B (en) * | 2019-06-19 | 2022-03-08 | 深圳大学 | Low-altitude near-real-time building earthquake damage assessment method |
| CN112232102A (en) * | 2019-07-15 | 2021-01-15 | 中国司法大数据研究院有限公司 | Building target identification method and system based on deep neural network and multitask learning |
| CN110516539A (en) * | 2019-07-17 | 2019-11-29 | 苏州中科天启遥感科技有限公司 | Remote sensing image building extracting method, system, storage medium and equipment based on confrontation network |
| CN110532894B (en) * | 2019-08-05 | 2021-09-03 | 西安电子科技大学 | Remote sensing target detection method based on boundary constraint CenterNet |
| CN110675408A (en) * | 2019-09-19 | 2020-01-10 | 成都数之联科技有限公司 | High-resolution image building extraction method and system based on deep learning |
| CN110674735B (en) * | 2019-09-23 | 2022-05-24 | 中国科学院地理科学与资源研究所 | Agricultural facility remote sensing extraction method and device based on fine classification |
| CN110738133B (en) * | 2019-09-23 | 2023-10-27 | 中科禾信遥感科技（苏州）有限公司 | Method and device for identifying image contour boundaries of different agricultural facilities |
| CN110717420A (en) * | 2019-09-25 | 2020-01-21 | 中国科学院深圳先进技术研究院 | Cultivated land extraction method and system based on remote sensing image and electronic equipment |
| CN110956196B (en) * | 2019-10-11 | 2024-03-08 | 东南大学 | Automatic recognition method for window wall ratio of urban building |
| CN111191500A (en) * | 2019-11-12 | 2020-05-22 | 广东融合通信股份有限公司 | Photovoltaic roof resource identification method based on deep learning image segmentation |
| CN110992257A (en) * | 2019-12-20 | 2020-04-10 | 北京航天泰坦科技股份有限公司 | Remote sensing image sensitive information automatic shielding method and device based on deep learning |
| CN111144335A (en) * | 2019-12-30 | 2020-05-12 | 自然资源部国土卫星遥感应用中心 | Method and device for building deep learning model |
| CN111242028A (en) * | 2020-01-13 | 2020-06-05 | 北京工业大学 | Remote sensing image ground object segmentation method based on U-Net |
| CN111339947A (en) * | 2020-02-26 | 2020-06-26 | 苏州中科天启遥感科技有限公司 | Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment |
| CN111368843B (en) * | 2020-03-06 | 2022-06-10 | 电子科技大学 | Method for extracting lake on ice based on semantic segmentation |
| CN111414878B (en) * | 2020-03-26 | 2023-11-03 | 北京京东智能城市大数据研究院 | Social attribute analysis and image processing method and device for land parcels |
| CN111489387B (en) * | 2020-04-09 | 2023-10-20 | 湖南盛鼎科技发展有限责任公司 | Remote sensing image building area calculation method |
| CN111476308A (en) * | 2020-04-10 | 2020-07-31 | 高崟 | Remote sensing image classification method and device based on prior geometric constraint and electronic equipment |
| CN111553272B (en) * | 2020-04-28 | 2022-05-06 | 上海市测绘院 | High-resolution satellite optical remote sensing image building change detection method based on deep learning |
| CN111914611B (en) * | 2020-05-09 | 2022-11-15 | 中国科学院空天信息创新研究院 | Urban green space high-resolution remote sensing monitoring method and system |
| CN111652251B (en) * | 2020-06-09 | 2023-06-27 | 星际空间（天津）科技发展有限公司 | Remote sensing image building feature extraction model construction method, device and storage medium |
| CN111652250B (en) * | 2020-06-09 | 2023-05-26 | 星际空间（天津）科技发展有限公司 | Remote sensing image building extraction method and device based on polygons and storage medium |
| CN111709346A (en) * | 2020-06-10 | 2020-09-25 | 嘉应学院 | Historical building identification and detection method based on deep learning and high-resolution images |
| CN111832502A (en) * | 2020-07-20 | 2020-10-27 | 中国人民解放军战略支援部队航天工程大学 | Remote sensing image visual salient region intelligent search method for satellite in-orbit application |
| CN111914328B (en) * | 2020-07-31 | 2023-11-17 | 同济大学 | Building plane aided design method and system based on artificial intelligence |
| CN112052783B (en) * | 2020-09-02 | 2024-04-09 | 中南大学 | High-resolution image weak supervision building extraction method combining pixel semantic association and boundary attention |
| CN112182125B (en) * | 2020-09-14 | 2022-07-05 | 中国科学院重庆绿色智能技术研究院 | Business gathering area boundary identification system |
| CN112149594B (en) * | 2020-09-29 | 2022-08-09 | 同济大学 | Urban construction assessment method based on deep learning and high-resolution satellite images |
| CN112288331B (en) * | 2020-11-24 | 2022-04-12 | 哈尔滨工业大学 | Building reconstruction evaluation system based on semantic recognition |
| CN112580510A (en) * | 2020-12-18 | 2021-03-30 | 广州市城市规划设计所 | Permeable ground area rate estimation method, permeable ground area rate estimation device, permeable ground area rate estimation equipment and storage medium |
| CN112990001A (en) * | 2021-03-11 | 2021-06-18 | 中国电建集团华东勘测设计研究院有限公司 | Remote sensing image monomer building rapid and accurate extraction method applied to digital collection |
| CN113516135B (en) * | 2021-06-23 | 2023-10-31 | 江苏师范大学 | Remote sensing image building extraction and contour optimization method based on deep learning |
| CN113505842B (en) * | 2021-07-21 | 2024-01-05 | 中国科学院空天信息创新研究院 | Automatic urban building extraction method suitable for large-scale regional remote sensing image |
| CN113486135B (en) * | 2021-07-27 | 2024-04-12 | 中南大学 | Building comprehensive method based on deep learning network |
| CN113610013A (en) * | 2021-08-10 | 2021-11-05 | 四川易方智慧科技有限公司 | Method for extracting building outline based on RGB (Red Green blue) wave bands of high-definition remote sensing image |
| CN113807301A (en) * | 2021-09-26 | 2021-12-17 | 武汉汉达瑞科技有限公司 | Automatic extraction method and automatic extraction system for newly-added construction land |
| CN114092813B (en) * | 2021-11-25 | 2022-08-05 | 中国科学院空天信息创新研究院 | Industrial park image extraction method and system, electronic equipment and storage medium |
| CN114331031B (en) * | 2021-12-08 | 2022-12-09 | 北京华清安地建筑设计有限公司 | Building traditional feature recognition and evaluation method and system |
| CN114049569B (en) * | 2022-01-13 | 2022-03-18 | 自然资源部第三地理信息制图院 | Deep learning model performance evaluation method and system |
| CN114677522B (en) * | 2022-03-17 | 2022-12-02 | 北京科技大学 | Building structure type prediction method and system based on building pattern spots and deep learning |
| CN114677589B (en) * | 2022-03-18 | 2022-12-06 | 广州市城市规划勘测设计研究院 | City functional area identification method, device, equipment and medium based on remote sensing interpretation |
| CN115841625B (en) * | 2023-02-23 | 2023-06-06 | 杭州电子科技大学 | Remote sensing building image extraction method based on improved U-Net model |
| CN115994987B (en) * | 2023-03-21 | 2023-06-06 | 天津市勘察设计院集团有限公司 | Rural building extraction and vectorization method based on inclined three-dimensional model |
| CN116052019B (en) * | 2023-03-31 | 2023-07-25 | 深圳市规划和自然资源数据管理中心(深圳市空间地理信息中心) | High-quality detection method suitable for built-up area of large-area high-resolution satellite image |
| CN117218485A (en) * | 2023-09-05 | 2023-12-12 | 安徽省第二测绘院 | Deep learning model-based multi-source remote sensing image interpretation sample library creation method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004922A (en) * | 2010-12-01 | 2011-04-06 | 南京大学 | High-resolution remote sensing image plane extraction method based on skeleton characteristic |
| CN104504718A (en) * | 2015-01-06 | 2015-04-08 | 南京大学 | High-definition aerial remote sensing data automatic road extraction method |
| CN107092877A (en) * | 2017-04-12 | 2017-08-25 | 武汉大学 | Remote sensing image roof contour extracting method based on basement bottom of the building vector |
| CN107247938A (en) * | 2017-06-08 | 2017-10-13 | 中国科学院遥感与数字地球研究所 | A kind of method of high-resolution remote sensing image City Building function classification |
| CN107767382A (en) * | 2017-09-26 | 2018-03-06 | 武汉市国土资源和规划信息中心 | The extraction method and system of static three-dimensional map contour of building line |
| CN107967454A (en) * | 2017-11-24 | 2018-04-27 | 武汉理工大学 | Take the two-way convolutional neural networks Classification in Remote Sensing Image method of spatial neighborhood relation into account |
| CN108198181A (en) * | 2018-01-23 | 2018-06-22 | 电子科技大学 | A kind of thermal-induced imagery processing method based on region segmentation and image co-registration |
| CN108537192A (en) * | 2018-04-17 | 2018-09-14 | 福州大学 | A kind of remote sensing image ground mulching sorting technique based on full convolutional network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9760990B2 (en) * | 2014-12-14 | 2017-09-12 | International Business Machines Corporation | Cloud-based infrastructure for feedback-driven training and image recognition |
| CA2995850A1 (en) * | 2015-08-31 | 2017-03-09 | Ryan Kottenstette | Systems and methods for analyzing remote sensing imagery |
-
2018
- 2018-10-30 CN CN201811279197.1A patent/CN109446992B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004922A (en) * | 2010-12-01 | 2011-04-06 | 南京大学 | High-resolution remote sensing image plane extraction method based on skeleton characteristic |
| CN104504718A (en) * | 2015-01-06 | 2015-04-08 | 南京大学 | High-definition aerial remote sensing data automatic road extraction method |
| CN107092877A (en) * | 2017-04-12 | 2017-08-25 | 武汉大学 | Remote sensing image roof contour extracting method based on basement bottom of the building vector |
| CN107247938A (en) * | 2017-06-08 | 2017-10-13 | 中国科学院遥感与数字地球研究所 | A kind of method of high-resolution remote sensing image City Building function classification |
| CN107767382A (en) * | 2017-09-26 | 2018-03-06 | 武汉市国土资源和规划信息中心 | The extraction method and system of static three-dimensional map contour of building line |
| CN107967454A (en) * | 2017-11-24 | 2018-04-27 | 武汉理工大学 | Take the two-way convolutional neural networks Classification in Remote Sensing Image method of spatial neighborhood relation into account |
| CN108198181A (en) * | 2018-01-23 | 2018-06-22 | 电子科技大学 | A kind of thermal-induced imagery processing method based on region segmentation and image co-registration |
| CN108537192A (en) * | 2018-04-17 | 2018-09-14 | 福州大学 | A kind of remote sensing image ground mulching sorting technique based on full convolutional network |
Non-Patent Citations (4)
| Title |
|---|
| A Boundary Regulated Network for Accurate Roof Segmentation and Outline Extraction;Guangming Wu等;《Remote Sening》;20180730;第10卷(第8期);第1-19页 * |
| Road Extraction From a High Spatial Resolution Remote Sensing Image Based on Richer Convolutional Features;ZHAOLI HONG等;《IEEE Access》;20180921;第6卷;第46988-47000页 * |
| 基于Ｕ型卷积神经网络的航空影像建筑物检测;伍光明等;《测绘学报》;20180630;第47卷(第6期);第864-872页 * |
| 基于深度学习U-net模型的高分辨率遥感影像分类方法研究;许慧敏;《中国优秀硕士学位论文全文数据库 基础科学辑》;20180915(第9期);第A008-39页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109446992A (en) | 2019-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109446992B (en) | Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN109493320B (en) | Remote sensing image road extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN110929607B (en) | Remote sensing identification method and system for urban building construction progress | |
| CN109543630B (en) | Remote sensing image woodland extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN103400151B (en) | The optical remote sensing image of integration and GIS autoregistration and Clean water withdraw method | |
| CN107358260B (en) | Multispectral image classification method based on surface wave CNN | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN110163213B (en) | Remote sensing image segmentation method based on disparity map and multi-scale depth network model | |
| CN112712033B (en) | Automatic division method for catchment areas of municipal drainage pipe network | |
| CN106295613A (en) | A kind of unmanned plane target localization method and system | |
| CN110570440A (en) | Image automatic segmentation method and device based on deep learning edge detection | |
| CN112906662B (en) | Method, device and equipment for detecting change of remote sensing image and storage medium | |
| Peeters et al. | Automated recognition of urban objects for morphological urban analysis | |
| CN113901900A (en) | Unsupervised change detection method and system for homologous or heterologous remote sensing image | |
| CN113657324A (en) | Urban functional area identification method based on remote sensing image ground object classification | |
| Yu et al. | Intelligent crack detection and quantification in the concrete bridge: A deep learning-assisted image processing approach | |
| CN112949407A (en) | Remote sensing image building vectorization method based on deep learning and point set optimization | |
| Zhu et al. | Change detection based on the combination of improved SegNet neural network and morphology | |
| CN114943902A (en) | Urban vegetation unmanned aerial vehicle remote sensing classification method based on multi-scale feature perception network | |
| CN112033937B (en) | Evaluation method for water extraction precision | |
| CN116704350B (en) | Water area change monitoring method and system based on high-resolution remote sensing image and electronic equipment | |
| CN106971402B (en) | SAR image change detection method based on optical assistance | |
| CN112613354A (en) | Heterogeneous remote sensing image change detection method based on sparse noise reduction self-encoder | |
| CN112084941A (en) | Target detection and identification method based on remote sensing image | |
| CN111079807A (en) | Ground object classification method and device |
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 |