CN111339947A - Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment - Google Patents
Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment Download PDFInfo
- Publication number
- CN111339947A CN111339947A CN202010120498.0A CN202010120498A CN111339947A CN 111339947 A CN111339947 A CN 111339947A CN 202010120498 A CN202010120498 A CN 202010120498A CN 111339947 A CN111339947 A CN 111339947A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing image
- vector
- boundary
- ground object
- 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
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000003860 storage Methods 0.000 title claims abstract description 6
- 239000013598 vector Substances 0.000 claims abstract description 92
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 41
- 230000011218 segmentation Effects 0.000 claims abstract description 36
- 238000007635 classification algorithm Methods 0.000 claims abstract description 21
- 238000013135 deep learning Methods 0.000 claims description 14
- 238000011156 evaluation Methods 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 230000004927 fusion Effects 0.000 claims description 9
- 238000012805 post-processing Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000009499 grossing Methods 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 4
- 230000009191 jumping Effects 0.000 claims description 3
- 238000002360 preparation method Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 abstract description 18
- 238000010586 diagram Methods 0.000 description 8
- 238000013527 convolutional neural network Methods 0.000 description 5
- 239000004576 sand Substances 0.000 description 5
- 239000003513 alkali Substances 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 244000025254 Cannabis sativa Species 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 2
- 241000209504 Poaceae Species 0.000 description 1
- FAPWRFPIFSIZLT-UHFFFAOYSA-M Sodium chloride Chemical compound [Na+].[Cl-] FAPWRFPIFSIZLT-UHFFFAOYSA-M 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 235000015097 nutrients Nutrition 0.000 description 1
- 230000000452 restraining effect Effects 0.000 description 1
- 239000011780 sodium chloride Substances 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
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
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Abstract
The invention provides a method for extracting a remote sensing image fuzzy boundary ground object, which comprises the following steps: acquiring a remote sensing image; carrying out classification extraction by using a supervision classification algorithm to obtain surface feature data; dividing by using a mean shift algorithm to obtain vector surface pattern spots; and fusing the ground feature data and the vector surface pattern spots to obtain the extracted remote sensing image fuzzy boundary ground feature data. The invention also relates to an extraction system, a storage medium and equipment of the remote sensing image fuzzy boundary ground object. The classification result and the segmentation result are fused to obtain the extracted fuzzy boundary ground feature data of the remote sensing image; according to the method, the ground boundary obtained by utilizing the automatic segmentation capability of the mean shift algorithm is finer, the form is more fit with the actual situation, so that the remote sensing image fuzzy boundary ground object is more accurately extracted and is more practical.
Description
Technical Field
The invention relates to the technical field of remote sensing image processing, in particular to a method for extracting a remote sensing image fuzzy boundary ground object.
Background
The remote sensing image features are complex and various, the representation of the same feature on the map can be very different, and the representation of different features on the map can be similar, which brings great challenges to the interpretation of the remote sensing image. Some land features on the remote sensing image are regular in shape, clear in boundary and easy to divide, such as buildings, roads and the like, some growing land features such as forest lands, grasslands and the like or land features greatly influenced by the growing land features such as sand lands, bare lands, saline-alkali lands and the like can greatly change along with time, seasons and other various reasons, such as form and sparseness degree, the form of the former is mainly determined by growth environment, weather factors and the like, the form of the latter is mainly determined by the growth vigor of the former, for example, the grasslands are degraded into the sand lands, and nutrient components of the bare lands disappear into the saline-alkali lands and the like. It is worth noting that the interiors of such land features, such as grasslands, have obvious differences due to different growth vigors, relatively independent geographical maps are formed, the sand lands adjacent to the deteriorated grasslands are relatively independent geographical maps, the geographical maps are usually distributed in a staggered mode, gradual transition exists between different categories and even between the same categories, and the boundaries are fuzzy, so that the weak boundaries of the remote sensing images are weak boundaries, and the extraction of the weak boundary land features is relatively difficult due to the fuzzy boundaries.
Common boundary extraction methods such as Sobel, Prewitt, Robert operators and the like mainly obtain boundaries through differences, and have a good extraction effect on strong boundary ground objects (with clear boundaries), but have a poor extraction effect on the condition that the interior of the ground objects or the space between the ground objects gradually changes or the boundary is fuzzy. In addition, although the canny operator is finer than the former extracted boundary, most of the image spot boundaries are not closed from the boundary extraction result, and are too fine and attractive to fit the actual situation. It can be confirmed that even the current powerful deep learning method cannot directly obtain the segmentation result of the boundary relatively conforming to the morphology of the fuzzy boundary ground object, because the key of the deep learning algorithm lies in the quality and data of the sample, but for the fuzzy boundary ground object, the boundary cannot be accurately drawn by naked eyes at all, and a certain deviation always exists.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a method for extracting the remote sensing image fuzzy boundary ground object. The invention realizes a shape sensitive fuzzy boundary extraction method by combining a shape segmentation and deep learning classification method using a mean shift algorithm.
The invention provides a method for extracting a remote sensing image fuzzy boundary ground object, which comprises the following steps:
acquiring a high-resolution remote sensing image, wherein a ground object to be extracted in the remote sensing image has a fuzzy boundary;
classifying and extracting the target area of the remote sensing image by using a supervision classification algorithm to obtain surface feature data of surface features to be extracted in the remote sensing image;
dividing the remote sensing image target area by using a mean shift algorithm to obtain mutually independent vector surface pattern spots;
and fusing the surface feature data and the vector surface pattern spots, and when the overlapping degree of the surface feature data and the vector surface pattern spots is larger than or equal to a set threshold range, giving the surface feature type of the surface feature data to the vector surface pattern spots to obtain the remote sensing image fuzzy boundary surface feature data.
Preferably, the classifying and extracting the remote sensing image target area by using a supervised classification algorithm includes:
the method for classifying and extracting the remote sensing image target area by using the deep learning classification algorithm comprises the following steps:
the method comprises the steps of segmenting and cutting a remote sensing image to obtain a surface vector file containing one or more fuzzy boundary ground objects, drawing ground object labels in the surface vector file, converting vector data of surface vector samples with a plurality of labels as the ground objects into raster data, and obtaining a rasterized ground object sample set;
adjusting UNET network model parameters, and performing UNET-based model training on the surface feature sample set to obtain a surface feature model;
inputting remote sensing image data to be tested into the ground object model, calculating a checking evaluation function of the test, and jumping to the next step if the evaluation function value reaches the standard; if the evaluation function value does not reach the standard, adjusting UNET network model parameters, returning to the step of sample preparation, modifying the sample set and carrying out iterative training again;
and performing surface feature prediction on a target area of the remote sensing image by using the surface feature model which is evaluated to reach the standard in the step precision evaluation to obtain the gridded surface feature data in the remote sensing image.
Preferably, the ground feature data and the vector surface pattern spots are subjected to broken spot removal and small hole filling processing.
Preferably, the step of segmenting the remote sensing image target region by using a mean shift algorithm to obtain mutually independent vector surface pattern spots further comprises:
and simplifying and smoothing the obtained vector surface pattern spots.
Preferably, the step of segmenting the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary comprises the following steps:
selecting a central point, and randomly selecting a point from the unclassified remote sensing image data as the central point;
acquiring a set, acquiring all points within a bandwidth from the central point, and recording as a set M, wherein the set M forms a cluster;
calculating vectors, calculating the vectors from the central point to each element in the set M, and adding the vectors to obtain an offset vector;
a drift movement, the center point moving along a drift direction, a distance of movement being a modulus of the offset vector;
obtaining a new central point, repeating the steps of obtaining a set, calculating a vector and moving in a drifting manner until the size of the offset vector is within a set threshold range, and obtaining the new central point;
forming a cluster, and repeating the steps of acquiring a set, calculating a vector and shifting until all points in the remote sensing image data are clustered;
and determining the cluster to which the current point belongs, namely the independent spot boundary, and taking the cluster with the maximum access frequency as the cluster to which the current point belongs according to the access frequency of each cluster to each point.
Preferably, in the step, the remote sensing image target area is segmented by using a mean shift algorithm to obtain mutually independent vector surface pattern spots, and the method includes:
dividing the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary; vectorizing the obtained independent image spot boundary, and performing surface construction to obtain the mutually independent vector surface image spots.
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 performing a method of extracting a remote sensing image blurred boundary feature.
A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program is used for executing the method for extracting the remote sensing image fuzzy boundary ground object by the processor.
A system for extracting remote sensing image fuzzy boundary ground objects comprises an acquisition module, a classification module, a segmentation module and a fusion module; wherein the content of the first and second substances,
the acquisition module is used for acquiring a high-resolution remote sensing image, and a ground object to be extracted in the remote sensing image has a fuzzy boundary;
the classification module is used for classifying and extracting the remote sensing image target area by using a supervision classification algorithm to obtain surface feature data of surface features to be extracted in the remote sensing image;
the segmentation module is used for segmenting the remote sensing image target area by using a mean shift algorithm to obtain mutually independent vector surface pattern spots;
and the fusion module is used for fusing the surface feature data and the vector surface pattern spots, and endowing the surface feature type of the surface feature data to the vector surface pattern spots when the overlapping degree of the surface feature data and the vector surface pattern spots is larger than or equal to a set threshold range, so as to obtain the remote sensing image fuzzy boundary surface feature data.
Preferably, the classification module comprises a deep learning classification algorithm unit, and the deep learning classification algorithm unit is used for classifying and extracting a remote sensing image target area to obtain the gridded ground feature data in the remote sensing image;
the segmentation module comprises a vectorization processing unit and a vectorization post-processing unit, wherein the vectorization processing unit is used for segmenting the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary; vectorizing the obtained independent image spot boundary, and performing surface construction to obtain the mutually independent vector surface image spots;
and the vectorization post-processing unit is used for simplifying and smoothing the obtained vector surface image spots.
Compared with the prior art, the invention has the beneficial effects that:
the invention discloses a method for extracting fuzzy boundary ground features of remote sensing images, which utilizes a network model to learn various characteristics of the ground features to obtain rough classification of the ground feature categories, and then utilizes a mean shift algorithm to extract the boundary of the same remote sensing image; the method comprises the steps of fusing a classification result and a segmentation result to obtain extracted remote sensing image fuzzy boundary ground feature data; according to the method, the ground boundary obtained by utilizing the automatic segmentation capability of the mean shift algorithm is finer, the form is more fit with the actual situation, so that the remote sensing image fuzzy boundary ground object is more accurately extracted and is more practical.
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 flowchart of an overall structure of a method for extracting a remote sensing image fuzzy boundary ground object according to the present invention;
FIG. 2 is a detailed flow chart of the supervised classification algorithm of the present invention;
FIG. 3 is a detailed flow chart of the mean segmentation algorithm of the present invention;
FIG. 4 is a schematic overall logic diagram of the method for extracting the remote sensing image fuzzy boundary ground object according to the present invention;
fig. 5 is a schematic diagram of the UNET network model of the present invention;
FIG. 6 is a logic diagram of the mean segmentation algorithm of the present invention;
FIG. 7 is a simplified schematic diagram of the fusion of terrain data and vector patches in accordance with the present invention;
FIG. 8 is a diagram of a remote sensing image target area according to the present invention;
FIG. 9 is a segmentation result of a target region using a mean shift algorithm according to the present invention;
FIG. 10 is an enlarged detail view of a portion of FIG. 9;
FIG. 11 is a schematic representation of grasses extracted using the UNET network model;
FIG. 12 is an enlarged detail view of a portion of FIG. 11 (the same area as FIG. 10);
FIG. 13 is a schematic view of FIG. 10 fused with FIG. 12;
FIG. 14 is a schematic representation of FIG. 13 after assigning grass categories thereto;
FIG. 15 is a schematic diagram illustrating the extraction of the fuzzy boundary ground object of the remote sensing image target area according to the present invention;
fig. 16 is a block diagram of an extraction system for a remote sensing image fuzzy boundary ground object according to 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 extraction of remote sensing image ground features (such as grassland and sand) is different from the conventional ground features (such as roads and buildings), and the boundary between the weak boundary ground features and other ground features with fuzzy boundaries like grassland, sand, forest land and the like is difficult to define, mainly because the generation or change of the ground features and the adjacent ground features are gradual and have no obvious boundary. In addition, the definition of the minimum geographical pattern of the remote sensing image, i.e. the minimum independent pattern at a certain scale, often appears in the same category. Taking the grassland as an example, the grassland with large area can not grow completely the same, and the grassland with the middle part grows sparsely, so that the sparse grassland is also an independent map spot relative to other grasslands and is divisible from the viewpoint of segmentation. Although the convolutional neural network is mature in semantic segmentation and widely applied to surface feature extraction at present, the segmentation capability of the surface features with fuzzy boundaries also greatly improves the space, because the growth distribution of the surface features is complex, especially the shape constructed by the edges is often not practical in vision, and the precision is natural.
The ground feature distribution of the remote sensing image has space complexity, and meanwhile, the intra-class difference and the inter-class difference of the ground features are also large, so that the classification problem of the ground features cannot be solved by using a method, and the fuzzy boundary ground features in the ground features are more difficult to extract by using a common method; the invention provides a classification and segmentation combined method for restraining the distribution of the fuzzy boundary ground objects, fully excavating the characteristics of the weak boundary ground objects and obtaining the pattern spots with beautiful shapes.
The invention provides a method for extracting a remote sensing image fuzzy boundary ground object, which comprises the following steps as shown in figures 1-7:
and S1, acquiring a high-resolution remote sensing image, wherein the ground object to be extracted in the remote sensing image has a fuzzy boundary. In one embodiment, a high-resolution No. 1 and 2-meter spatial resolution 4-waveband remote sensing image is acquired, and one or more than one type of land features with fuzzy boundaries, such as grasslands, sandy land, saline land and the like, are selected.
And S2, classifying and extracting the target area of the remote sensing image by using a supervision classification algorithm to obtain the feature data of the feature to be extracted in the remote sensing image. In one embodiment, supervised classification is one type of remote sensing image classification, i.e., a process of identifying pixels of other unknown classes using sample pixels of the identified class. The sample pixels of the confirmed category refer to those located in a training area, in the category, an analyst selects a certain number of training areas for each category on an image, a computer calculates statistics or other information of each training sample area, each pixel is compared with a training sample, and the pixel is classified into a sample class most similar to the training sample according to different rules. Supervised classification can be divided into two basic steps: selecting training samples and extracting statistical information, and selecting a classification algorithm. The fuzzy boundary ground object classes are roughly extracted by using the deep learning classification model, and the extraction effect is more stable by using the deep learning method compared with the traditional method.
Specifically, the classification and extraction of the remote sensing image target area by using a supervised classification algorithm includes:
the remote sensing image target area is classified and extracted by using a deep learning classification algorithm, as shown in fig. 2, the method comprises the following steps:
s21, the remote sensing image is divided and cut to obtain a face vector file containing one or more fuzzy boundary ground objects, ground object labels in the face vector file are drawn, vector data of the face vector samples with the labels as the ground objects are converted into raster data, and a rasterized ground object sample set is obtained. And selecting a small amount of 1000 x 1000 samples from the obtained high-resolution No. 1 and 2 m spatial resolution 4-waveband remote sensing image, and drawing and converting to obtain a sample set containing images and labels.
And S22, adjusting UNET network model parameters, and performing UNET-based model training on the feature sample set to obtain a feature model. As shown in fig. 5, the sample set is input into a convolutional neural network, and the convolutional neural network model is preferably UNET (full convolutional neural network image segmentation), and UNET is a convolutional neural network which is obtained by adding an upsampling stage on the basis of an FCN (full convolutional network), has multiple scales, and is suitable for remote sensing large atlas semantic segmentation. The UNET network structure is mainly divided into two parts, the left part is a convolution network for extracting image features like the FCN, the right part is deconvolution operation of an up-sampling part, and a plurality of feature channels are added at the stage, so that more original image texture information is allowed to be transmitted in a high-resolution layer. Particularly, UNET has no fully connected layer, and the whole process uses valid to perform convolution, so that it can be ensured that the segmentation result is obtained based on the context features that are not missing, and therefore, the image sizes of input and output are inconsistent, and for the input of a large image, the overlap-progression can be used to perform seamless image output. In addition, the up-sampling part fuses the output of the feature extraction part, so that the multi-scale fusion is realized, the feature of the object is the output from the first convolution block and the up-sampling output, and the connection runs through the whole network, so that the image feature can be fully extracted.
It should be noted that the rough classification result is obtained through the UNET network model, and a fine classification result is not required to be obtained; general network models are improved and optimized by referring to the UNET model, so that the UNET network model can meet the requirements in the embodiment by obtaining results through parameter adjustment.
S23, inputting the remote sensing image data to be tested into the ground feature model, calculating the checking evaluation function of the test, and jumping to the next step if the evaluation function value reaches the standard; and if the evaluation function value does not reach the standard, adjusting parameters of the UNET network model, returning to the step of sample preparation, modifying the sample set and carrying out iterative training again. The evaluation function preferably selects an MIoU (mean intersection-over-intersection) value and an MPA (mean pixel access), if the MIoU value and the MPA value both reach the standard, the next step is carried out, if the MIoU value or the MPA value do not reach the standard, the UNET network model parameter is adjusted, the UNET network model parameter is returned to S1, and the sample is modified for iterative training again; for example, if the MIoU value is set to be in the range of 0.70-1, the MPA value is set to be in the range of 0.75-1, the UNET network model parameters are adjusted and returned to S21 for retraining when the range is exceeded.
And S24, performing surface feature prediction on the target area of the remote sensing image by using the surface feature model which is evaluated to reach the standard in the step precision evaluation to obtain the grid surface feature data in the remote sensing image. Inputting the target area of the remote sensing image into the ground object model which is evaluated to reach the standard in the step S23 to obtain a classification result, and performing post-processing on the raster result, wherein the post-processing of the raster mainly performs broken spot removal and small hole filling on the classification result to ensure the attractiveness of the result.
And S3, segmenting the remote sensing image target area by using a mean shift algorithm to obtain mutually independent vector surface pattern spots. In one embodiment, the same remote sensing image target area as in step S2 is segmented using a mean shift algorithm. The mean shift algorithm has automatic segmentation capability, does not need to set parameters manually, and has better effect than the common clustering method. The mean shift algorithm is based on a sliding window, a data point dense region is iterated, most importantly, compared with other clustering algorithms, the number of categories is determined through iteration instead of being set, and the algorithm mainly focuses on segmentation to obtain reasonable edges and the categories are given through a convolutional network.
The basic formula of the mean shift algorithm is
ShIn xiA high-dimensional sphere area with the radius of h as a central point; k is contained in ShThe number of points within the range; x is the number ofiIs contained in ShPoints within the range are used to calculate the mean of the drift. The center update formula is xt+1＝Mt+xtFor shifting the center pointMoving to an offset mean value position, wherein Mt is the offset mean value obtained under the state of t; xt is the center in the t state.
Specifically, as shown in fig. 3 and 6, the method includes the following steps:
s31, selecting a central point, and randomly selecting a point from the unclassified remote sensing image data as the central point;
s32, acquiring a set, acquiring all points within the bandwidth from the central point, and recording as a set M, wherein the set M forms a cluster;
s33, calculating vectors, namely calculating the vectors from the central point to each element in the set M, and adding the vectors to obtain an offset vector;
s34, shifting, wherein the central point shifts along the shifting direction, and the shifting distance is the mode of the offset vector;
s35, obtaining a new central point, repeating the steps of obtaining a set, calculating vectors and shifting until the size of the offset vector is within a set threshold range, and obtaining the new central point at the moment;
s36, forming clusters, repeating the steps to obtain a set, calculate vectors and shift movement until all points in the remote sensing image data are clustered;
and S37, determining the cluster to which the current point belongs, and taking the cluster with the maximum access frequency as the cluster to which the current point belongs, namely the independent spot boundary according to the access frequency of each cluster to each point.
The method further comprises the following steps after the step S37:
s38, segmenting the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary; vectorizing the obtained independent image spot boundary, and performing surface construction to obtain the mutually independent vector surface image spots. In one embodiment, a remote sensing image target area is subjected to multi-scale segmentation by using a mean shift algorithm to obtain a boundary attached with an independent image spot, so that the morphological best state is achieved. Vectorizing the obtained independent image spots, and performing extension and surface construction processing on the isolated line segments to obtain mutually independent vector surface image spots. And carrying out grid processing on the vector surface pattern spots obtained by the mean shift algorithm, namely, carrying out broken spot removal and small hole filling, and ensuring the attractiveness of the result.
It should be noted that the mean shift algorithm obtains the edge morphology of the minimum inseparable pattern spot of the remote sensing image under a certain scale, and when the obtained pattern spot is not fine enough, the segmentation scale can be adjusted to obtain a more detailed pattern spot morphology.
S39, vector post-processing, the vector surface patches obtained are severely jagged, and it is necessary to simplify and smooth the vector surface patches to obtain beautiful vector surface patches with little change in overall shape.
And S4, fusing the feature data and the vector surface pattern spots, and when the overlap degree of the feature data and the vector surface pattern spots is larger than or equal to a set threshold range, giving the feature type of the feature data to the vector surface pattern spots to obtain the remote sensing image fuzzy boundary feature data. In one embodiment, as shown in FIG. 7, the terrain data obtained in step S2 is fused with the vector surface map patches obtained in step S3; generally, the area size of each image spot is calculated, when the coincidence degree of the obtained terrain data and the area of the obtained vector surface image spot is larger than or equal to a set threshold range, the threshold range is generally set to be 70% -90%, that is, when the coincidence degree of the terrain data obtained through the network model and the vector surface image spot at the same position obtained through the mean shift algorithm is within 70% -90%, the boundary of the vector surface image spot obtained through the mean shift algorithm is an extracted terrain boundary and is given to the terrain type of the vector surface image spot, for example, the terrain data at the position obtained through the network model is a grassland, the grassland is given to the vector surface image spot, and finally, the extraction result of the terrain with the shape conforming to the actual fuzzy boundary terrain is obtained.
For example, as shown in fig. 8-15, taking grass as an example, high density grass patches are found and assigned by overlapping degree statistics. Through actual production, for 169 square kilometers of a target area, the total time for extracting all weak boundary ground objects including grasslands, saline-alkali soil, bare land and the like is only about 4 hours. The manual drawing is time-consuming, and distinguishing the minimum separable image spots of the remote sensing images under different scales by naked eyes is quite difficult, so that manual drawing cannot be implemented basically. The method effectively improves the efficiency of extracting the ground objects of the fuzzy boundary of the remote sensing image under the condition of ensuring the precision. Fig. 8 shows a four-band remote sensing image with weak boundary ground objects as the main target area of the remote sensing image of the present invention, including grasslands, bare lands, saline-alkali lands, etc., where the grasslands are distributed more complicated and the sparsity of different areas is greatly different; FIG. 9 is a segmentation result of the entire target region by the mean shift algorithm, which can reduce the degree of segmentation of the remote sensing image to the smallest inseparable image spot at the set segmentation scale; fig. 10 is an enlarged view, i.e. a detail display diagram, of fig. 9, and it can be seen that the algorithm obtains the edge morphology of the minimum inseparable patch of the remote sensing image under a certain scale, whether between categories or within categories. In addition, if the image spots are not fine enough, the segmentation scale can be adjusted to obtain a more detailed image spot form; FIG. 11 is a classification result of the UNET network model, and the test takes high-density grassland as an example, and the high-density grassland of the whole target area is extracted through the network model; fig. 12 is a detail display of the UNET network model on the high-density grassland classification result, where the land boundary is not yet fit to the actual boundary, and needs to be optimized in combination with the segmentation result; FIG. 13 is a result of fusion of similar patches after combining the segmentation result of the mean shift algorithm and the classification result of the weak boundary ground object, and a result obtained by fusing the segmentation patches of the same category after fusing the segmentation results with the categories of grassland, sandy land, bare land, and the like; FIG. 14 is a result of assigning classes to segmented patches by calculation of overlap to high density meadow gathers; fig. 15 shows the extraction result of the feature with the fuzzy boundary in the target area of the remote sensing image.
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 performing a method of extracting a remote sensing image blurred boundary feature. A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program is used for executing the method for extracting the remote sensing image fuzzy boundary ground object by the processor.
A system for extracting remote sensing image fuzzy boundary ground objects is shown in figure 16 and comprises an acquisition module, a classification module, a segmentation module and a fusion module; wherein the content of the first and second substances,
the acquisition module is used for acquiring a high-resolution remote sensing image, and a ground object to be extracted in the remote sensing image has a fuzzy boundary;
the classification module is used for classifying and extracting the remote sensing image target area by using a supervision classification algorithm to obtain surface feature data of surface features to be extracted in the remote sensing image;
the segmentation module is used for segmenting the remote sensing image target area by using a mean shift algorithm to obtain mutually independent vector surface pattern spots;
and the fusion module is used for fusing the surface feature data and the vector surface pattern spots, and endowing the surface feature type of the surface feature data to the vector surface pattern spots when the overlapping degree of the surface feature data and the vector surface pattern spots is larger than or equal to a set threshold range, so as to obtain the remote sensing image fuzzy boundary surface feature data.
Further, the classification module comprises a deep learning classification algorithm unit, and classification extraction is carried out on a remote sensing image target area by using the deep learning classification algorithm unit to obtain the gridded ground feature data in the remote sensing image;
the segmentation module comprises a vectorization processing unit and a vectorization post-processing unit, wherein the vectorization processing unit is used for segmenting the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary; vectorizing the obtained independent image spot boundary, and performing surface construction to obtain the mutually independent vector surface image spots;
and the vectorization post-processing unit is used for simplifying and smoothing the obtained vector surface image spots.
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. A method for extracting a remote sensing image fuzzy boundary ground object is characterized by comprising the following steps:
acquiring a high-resolution remote sensing image, wherein a ground object to be extracted in the remote sensing image has a fuzzy boundary;
classifying and extracting the target area of the remote sensing image by using a supervision classification algorithm to obtain surface feature data of surface features to be extracted in the remote sensing image;
dividing the remote sensing image target area by using a mean shift algorithm to obtain mutually independent vector surface pattern spots;
and fusing the surface feature data and the vector surface pattern spots, and when the overlapping degree of the surface feature data and the vector surface pattern spots is larger than or equal to a set threshold range, giving the surface feature type of the surface feature data to the vector surface pattern spots to obtain the remote sensing image fuzzy boundary surface feature data.
2. The method for extracting the remote sensing image fuzzy boundary ground object as claimed in claim 1, wherein the classifying and extracting the remote sensing image target area by using the supervised classification algorithm comprises:
the method for classifying and extracting the remote sensing image target area by using the deep learning classification algorithm comprises the following steps:
the method comprises the steps of segmenting and cutting a remote sensing image to obtain a surface vector file containing one or more fuzzy boundary ground objects, drawing ground object labels in the surface vector file, converting vector data of surface vector samples with a plurality of labels as the ground objects into raster data, and obtaining a rasterized ground object sample set;
adjusting UNET network model parameters, and performing UNET-based model training on the surface feature sample set to obtain a surface feature model;
inputting remote sensing image data to be tested into the ground object model, calculating a checking evaluation function of the test, and jumping to the next step if the evaluation function value reaches the standard; if the evaluation function value does not reach the standard, adjusting UNET network model parameters, returning to the step of sample preparation, modifying the sample set and carrying out iterative training again;
and performing surface feature prediction on a target area of the remote sensing image by using the surface feature model which is evaluated to reach the standard in the step precision evaluation to obtain the gridded surface feature data in the remote sensing image.
3. The method for extracting the remote sensing image fuzzy boundary ground object as claimed in claim 2, wherein the ground object data and the vector surface pattern spot are subjected to broken spot removal and small hole filling processing.
4. The method for extracting the remote sensing image fuzzy boundary ground object according to claim 1, wherein the step of segmenting the remote sensing image target area by using a mean shift algorithm further comprises the following steps of after mutually independent vector surface pattern spots are obtained:
and simplifying and smoothing the obtained vector surface pattern spots.
5. The method for extracting the remote sensing image fuzzy boundary ground object as claimed in claim 1, wherein the step of segmenting the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary comprises the following steps:
selecting a central point, and randomly selecting a point from the unclassified remote sensing image data as the central point;
acquiring a set, acquiring all points within a bandwidth from the central point, and recording as a set M, wherein the set M forms a cluster;
calculating vectors, calculating the vectors from the central point to each element in the set M, and adding the vectors to obtain an offset vector;
a drift movement, the center point moving along a drift direction, a distance of movement being a modulus of the offset vector;
obtaining a new central point, repeating the steps of obtaining a set, calculating a vector and moving in a drifting manner until the size of the offset vector is within a set threshold range, and obtaining the new central point;
forming a cluster, and repeating the steps of acquiring a set, calculating a vector and shifting until all points in the remote sensing image data are clustered;
and determining the cluster to which the current point belongs, namely the independent spot boundary, and taking the cluster with the maximum access frequency as the cluster to which the current point belongs according to the access frequency of each cluster to each point.
6. The method for extracting the remote sensing image fuzzy boundary ground object according to claim 1, wherein the step of segmenting the remote sensing image target area by using a mean shift algorithm to obtain mutually independent vector surface pattern spots comprises the following steps:
dividing the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary; vectorizing the obtained independent image spot boundary, and performing surface construction to obtain the mutually independent vector surface image spots.
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. A system for extracting a remote sensing image fuzzy boundary ground object is characterized by comprising an acquisition module, a classification module, a segmentation module and a fusion module; wherein the content of the first and second substances,
the acquisition module is used for acquiring a high-resolution remote sensing image, and a ground object to be extracted in the remote sensing image has a fuzzy boundary;
the classification module is used for classifying and extracting the remote sensing image target area by using a supervision classification algorithm to obtain surface feature data of surface features to be extracted in the remote sensing image;
the segmentation module is used for segmenting the remote sensing image target area by using a mean shift algorithm to obtain mutually independent vector surface pattern spots;
and the fusion module is used for fusing the surface feature data and the vector surface pattern spots, and endowing the surface feature type of the surface feature data to the vector surface pattern spots when the overlapping degree of the surface feature data and the vector surface pattern spots is larger than or equal to a set threshold range, so as to obtain the remote sensing image fuzzy boundary surface feature data.
10. The system for extracting the remote sensing image fuzzy boundary ground object as claimed in claim 9, wherein said classification module comprises a deep learning classification algorithm unit, said deep learning classification algorithm unit is used for classifying and extracting the remote sensing image target area to obtain the grid ground object data in the remote sensing image;
the segmentation module comprises a vectorization processing unit and a vectorization post-processing unit, wherein the vectorization processing unit is used for segmenting the remote sensing image target area by using a mean shift algorithm to obtain an independent image spot boundary; vectorizing the obtained independent image spot boundary, and performing surface construction to obtain the mutually independent vector surface image spots;
and the vectorization post-processing unit is used for simplifying and smoothing the obtained vector surface image spots.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010120498.0A CN111339947A (en) | 2020-02-26 | 2020-02-26 | Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010120498.0A CN111339947A (en) | 2020-02-26 | 2020-02-26 | Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111339947A true CN111339947A (en) | 2020-06-26 |
Family
ID=71185708
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010120498.0A Pending CN111339947A (en) | 2020-02-26 | 2020-02-26 | Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111339947A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111784606A (en) * | 2020-06-30 | 2020-10-16 | 长江大学 | Remote sensing image classification post-processing method, storage medium and system |
| CN112084871A (en) * | 2020-08-10 | 2020-12-15 | 浙江工业大学 | High-resolution remote sensing target boundary extraction method based on weak supervised learning |
| CN112435274A (en) * | 2020-11-09 | 2021-03-02 | 国交空间信息技术（北京）有限公司 | Remote sensing image planar ground object extraction method based on object-oriented segmentation |
| CN112581468A (en) * | 2020-12-29 | 2021-03-30 | 二十一世纪空间技术应用股份有限公司 | Processing method and device for extracting information facing remote sensing image |
| CN113822900A (en) * | 2021-07-09 | 2021-12-21 | 武汉大学 | Vector constraint object-oriented new image sample automatic selection method and system |
| CN114067221A (en) * | 2022-01-14 | 2022-02-18 | 成都数联云算科技有限公司 | Remote sensing image woodland extraction method, system, device and medium |
| CN114708222A (en) * | 2022-04-02 | 2022-07-05 | 广西壮族自治区自然资源遥感院 | Remote sensing image change detection quality evaluation method based on target area distribution characteristics |
| CN115578607A (en) * | 2022-12-08 | 2023-01-06 | 自然资源部第三航测遥感院 | Method for rapidly extracting coverage area of effective pixels of remote sensing image |
| CN116047546A (en) * | 2022-07-07 | 2023-05-02 | 北京玖天气象科技有限公司 | Mountain fire monitoring method based on multi-source satellite data |
| CN117274814A (en) * | 2023-10-08 | 2023-12-22 | 北京香田智能科技有限公司 | Tobacco field image processing method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109446992A (en) * | 2018-10-30 | 2019-03-08 | 苏州中科天启遥感科技有限公司 | Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning |
-
2020
- 2020-02-26 CN CN202010120498.0A patent/CN111339947A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109446992A (en) * | 2018-10-30 | 2019-03-08 | 苏州中科天启遥感科技有限公司 | Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning |
Non-Patent Citations (5)
| Title |
|---|
| LZHF1122: "均值漂移", pages 4, Retrieved from the Internet <URL:lzhf1122，http://t.csdnimg.cn/O8cdg> * |
| 刘纯;洪亮;陈杰;楚森森;邓敏;: "融合像素―多尺度区域特征的高分辨率遥感影像分类算法", 遥感学报, no. 02, 25 March 2015 (2015-03-25) * |
| 方旭;王光辉;杨化超;刘慧杰;闫立波;: "结合均值漂移分割与全卷积神经网络的高分辨遥感影像分类", 激光与光电子学进展, no. 02, 20 September 2017 (2017-09-20), pages 1 - 3 * |
| 楚森森;洪亮;陈杰;邓敏;杨昆;刘纯;: "融合边界信息的高分辨率遥感影像分割优化算法", 中国图象图形学报, no. 08, 16 August 2016 (2016-08-16) * |
| 邓刘洋;沈占锋;柯映明;: "城市建成区遥感影像边界提取与扩张分析", 地球信息科学学报, no. 07, 11 July 2018 (2018-07-11) * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111784606B (en) * | 2020-06-30 | 2023-11-24 | 长江大学 | Remote sensing image classification post-processing method, storage medium and system |
| CN111784606A (en) * | 2020-06-30 | 2020-10-16 | 长江大学 | Remote sensing image classification post-processing method, storage medium and system |
| CN112084871A (en) * | 2020-08-10 | 2020-12-15 | 浙江工业大学 | High-resolution remote sensing target boundary extraction method based on weak supervised learning |
| CN112084871B (en) * | 2020-08-10 | 2024-02-13 | 浙江工业大学 | High-resolution remote sensing target boundary extraction method based on weak supervised learning |
| CN112435274A (en) * | 2020-11-09 | 2021-03-02 | 国交空间信息技术（北京）有限公司 | Remote sensing image planar ground object extraction method based on object-oriented segmentation |
| CN112581468A (en) * | 2020-12-29 | 2021-03-30 | 二十一世纪空间技术应用股份有限公司 | Processing method and device for extracting information facing remote sensing image |
| CN112581468B (en) * | 2020-12-29 | 2023-09-22 | 二十一世纪空间技术应用股份有限公司 | Processing method and device for extracting information for remote sensing image |
| CN113822900A (en) * | 2021-07-09 | 2021-12-21 | 武汉大学 | Vector constraint object-oriented new image sample automatic selection method and system |
| CN113822900B (en) * | 2021-07-09 | 2023-09-22 | 武汉大学 | Method and system for automatically selecting new image sample based on vector constraint object-oriented |
| CN114067221A (en) * | 2022-01-14 | 2022-02-18 | 成都数联云算科技有限公司 | Remote sensing image woodland extraction method, system, device and medium |
| CN114708222A (en) * | 2022-04-02 | 2022-07-05 | 广西壮族自治区自然资源遥感院 | Remote sensing image change detection quality evaluation method based on target area distribution characteristics |
| CN116047546A (en) * | 2022-07-07 | 2023-05-02 | 北京玖天气象科技有限公司 | Mountain fire monitoring method based on multi-source satellite data |
| CN116047546B (en) * | 2022-07-07 | 2024-02-27 | 北京玖天气象科技有限公司 | Mountain fire monitoring method based on multi-source satellite data |
| CN115578607A (en) * | 2022-12-08 | 2023-01-06 | 自然资源部第三航测遥感院 | Method for rapidly extracting coverage area of effective pixels of remote sensing image |
| CN117274814A (en) * | 2023-10-08 | 2023-12-22 | 北京香田智能科技有限公司 | Tobacco field image processing method |
| CN117274814B (en) * | 2023-10-08 | 2024-02-20 | 北京香田智能科技有限公司 | Tobacco field image processing method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111339947A (en) | Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment | |
| US10984532B2 (en) | Joint deep learning for land cover and land use classification | |
| CN108573276B (en) | Change detection method based on high-resolution remote sensing image | |
| CN108830870B (en) | Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning | |
| Blaschke et al. | Object-oriented image analysis and scale-space: theory and methods for modeling and evaluating multiscale landscape structure | |
| EP3614308A1 (en) | Joint deep learning for land cover and land use classification | |
| Wang et al. | An automatic region-based image segmentation algorithm for remote sensing applications | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN101840581B (en) | Method for extracting profile of building from satellite remote sensing image | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN105069796B (en) | SAR image segmentation method based on small echo both scatternets | |
| Karami et al. | Gully erosion mapping using object-based and pixel-based image classification methods | |
| CN104102928B (en) | A kind of Classifying Method in Remote Sensing Image based on texture primitive | |
| CN105138970A (en) | Spatial information-based polarization SAR image classification method | |
| CN104036294A (en) | Spectral tag based adaptive multi-spectral remote sensing image classification method | |
| Skidmore | Unsupervised training area selection in forests using a nonparametric distance measure and spatial information | |
| Li et al. | An efficient multi-scale segmentation for high-resolution remote sensing imagery based on statistical region merging and minimum heterogeneity rule | |
| CN109344837B (en) | SAR image semantic segmentation method based on deep convolutional network and weak supervised learning | |
| Albert et al. | Land use classification using conditional random fields for the verification of geospatial databases | |
| Mohammadzadeh et al. | Automatic main road extraction from high resolution satellite imageries by means of particle swarm optimization applied to a fuzzy-based mean calculation approach | |
| Forghani et al. | Extracting terrain categories from multi-source satellite imagery | |
| CN114862883A (en) | Target edge extraction method, image segmentation method and system | |
| Hashim et al. | Multi-level image segmentation for urban land-cover classifications | |
| Candare et al. | Mapping of high value crops through an object-based SVM model using lidar data and orthophoto in Agusan del Norte Philippines | |
| Djerriri et al. | Object-based classification of Sentinel-2 imagery using compact texture unit descriptors through Google Earth Engine |
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 |