CN114283286A - Remote sensing image segmentation method and device and electronic equipment - Google Patents
Remote sensing image segmentation method and device and electronic equipment Download PDFInfo
- Publication number
- CN114283286A CN114283286A CN202111665954.0A CN202111665954A CN114283286A CN 114283286 A CN114283286 A CN 114283286A CN 202111665954 A CN202111665954 A CN 202111665954A CN 114283286 A CN114283286 A CN 114283286A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing image
- target element
- extraction
- sample
- 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 48
- 238000003709 image segmentation Methods 0.000 title claims abstract description 38
- 238000000605 extraction Methods 0.000 claims abstract description 106
- 238000012549 training Methods 0.000 claims description 56
- 230000011218 segmentation Effects 0.000 claims description 14
- 238000005070 sampling Methods 0.000 claims description 13
- 230000002194 synthesizing effect Effects 0.000 claims description 7
- 238000010586 diagram Methods 0.000 description 10
- 238000001228 spectrum Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000007636 ensemble learning method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000013507 mapping 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
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a remote sensing image segmentation method, a remote sensing image segmentation device and electronic equipment, wherein the method comprises the following steps: the method comprises the steps of obtaining a remote sensing image to be segmented, extracting target elements for multiple times from the remote sensing image to obtain multiple extraction results, then integrating the multiple extraction results to obtain a final target element extraction result, improving the accuracy of the final target element extraction result, and further based on the final target element extraction result, accurately segmenting the target elements from the remote sensing image to achieve the purpose of improving the accuracy of the remote sensing image segmentation result.
Description
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a method and an apparatus for segmenting a remote sensing image, and an electronic device.
Background
The remote sensing image segmentation is to classify each pixel in the remote sensing image, the pixels with the same value are classified into the same class, and the method can be used for practical application of typical ground object analysis, road extraction, city planning and the like, and has great significance to civil affairs and military affairs. Earlier, remote sensing images were typically segmented using conventional methods, such as: threshold-based segmentation methods, region-based segmentation methods, edge-based segmentation methods, etc., but these methods can only extract low-level features. With the development of remote sensing technology, high-resolution remote sensing images are easier to obtain, and meanwhile, the detail characteristics of the remote sensing images are more and more complex. The traditional remote sensing image segmentation method is adopted to segment the high-resolution remote sensing image, the accuracy of the obtained image segmentation result is low, and the segmentation requirement of the high-resolution remote sensing image cannot be met.
Disclosure of Invention
In view of this, the present disclosure provides a method and an apparatus for segmenting a remote sensing image, and an electronic device, which can solve the problem of low accuracy of a segmentation result of a remote sensing image.
In order to solve the technical problem, the present disclosure is implemented as follows:
according to an aspect of the disclosure, there is provided a remote sensing image segmentation method, including:
acquiring a remote sensing image to be segmented;
extracting target elements for multiple times from the remote sensing image to obtain multiple extraction results, and synthesizing the multiple extraction results to obtain a final target element extraction result;
and based on the final target element extraction result, dividing the target element in the remote sensing image.
In a possible implementation mode, extracting the target elements for multiple times from the remote sensing image to obtain multiple extraction results, and synthesizing the multiple extraction results to obtain a final target element extraction result, wherein the extraction is performed through a pre-constructed strong classifier;
and the strong classifier is constructed by adopting an integrated learning algorithm.
In one possible implementation, constructing the strong classifier using an ensemble learning algorithm includes:
acquiring a training data set and at least two basic weak classifiers;
respectively training at least two basic weak classifiers based on the training data set to obtain at least two weak classifiers;
and for at least two weak classifiers, obtaining the strong classifier through a preset combination strategy.
In one possible implementation, the sample data in the training data set is constructed by:
selecting a sample remote sensing image;
drawing vector data of target elements in the same coordinate system with the sample remote sensing image according to a target extraction task;
converting the vector data of the target element into raster data with the size consistent with that of the sample remote sensing image, and taking the raster data as label data corresponding to the sample remote sensing image;
and obtaining the sample data according to the sample remote sensing image and the label data.
In one possible implementation, training the at least two basic weak classifiers based on the training data set includes:
extracting a training subset corresponding to the basic weak classifier from the training data set by a replaced extraction method;
and training the basic weak classifiers through the training subsets corresponding to the basic weak classifiers to obtain the weak classifiers.
In one possible implementation, the weak classifier is implemented using a U-Net network.
In one possible implementation manner, the U-Net network comprises a six-layer down-sampling coding layer and a six-layer up-sampling decoding layer which are sequentially cascaded.
In one possible implementation, each of the up-sampling decoding layers includes a channel space attention module;
the channel space attention module is cascaded after the last convolution layer of each up-sampling decoding layer.
According to a second aspect of the present disclosure, there is provided a remote sensing image segmentation apparatus comprising:
the image acquisition module is used for acquiring a remote sensing image to be segmented;
the target extraction module is used for extracting target elements from the remote sensing image for multiple times to obtain multiple extraction results, and synthesizing the multiple extraction results to obtain a final target element extraction result;
and the target segmentation module is used for segmenting the target element in the remote sensing image based on the final target element extraction result.
According to an aspect of the present disclosure, there is provided a remote sensing image segmentation electronic device including:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to carry out the executable instructions when implementing the method of any one of the first aspects of the present disclosure.
In the method, the remote sensing image to be segmented is obtained, multiple times of target element extraction are carried out on the remote sensing image to obtain multiple extraction results, and then the multiple extraction results are integrated to obtain a final target element extraction result, so that even if one or a small number of extraction results in the multiple extraction results have deviation, the multiple extraction results can be corrected in a mode of integrating the multiple extraction results, the accuracy of the final target element extraction result is improved, and further, the target elements can be accurately segmented in the remote sensing image based on the final target element extraction result, and the purpose of improving the accuracy of the remote sensing image segmentation result is achieved.
Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.
FIG. 1 shows a schematic flow diagram of a method of remote sensing image segmentation according to an embodiment of the present disclosure;
FIG. 2 shows a schematic structural diagram of a U-Net network according to an embodiment of the present disclosure;
FIG. 3 shows a schematic flow diagram of an example of a remote sensing image segmentation method according to an embodiment of the present disclosure;
FIG. 4 is a diagram illustrating a comparison of remote sensing image segmentation method with other network structure segmentation results according to an embodiment of the present disclosure;
FIG. 5 shows a schematic block diagram of a remote sensing image segmentation apparatus according to an embodiment of the present disclosure;
FIG. 6 shows a schematic block diagram of remote sensing image segmentation electronics, in accordance with an embodiment of the present disclosure.
Detailed Description
Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.
The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.
Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better understanding of the present disclosure. It will be understood by those skilled in the art that the present disclosure may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present disclosure.
< method examples >
FIG. 1 shows a schematic flow diagram of a method of remote sensing image segmentation according to an embodiment of the present disclosure.
As shown in fig. 1, the method includes steps S110 to S130.
And S110, acquiring a remote sensing image to be segmented.
When analyzing the target elements of the set area, the remote sensing image of the set area can be acquired as the remote sensing image to be segmented, so as to segment the target elements of interest in the remote sensing image to be segmented. The target elements may be surface feature elements in the remote sensing image, such as: roads, buildings, vehicles, etc., and may also be other analytical elements of interest, and are not specifically limited herein. The set area and the target element can be selected in advance according to the application scene. For example, when a building in area a is planned, area a may be selected as a set area, and the building may be selected as a target element. For another example, when planning a road in the B area, the B area may be selected as a set area, and the road may be selected as a target element.
The remote sensing image to be segmented may be a historical remote sensing image, or a remote sensing image acquired in real time by a remote sensing satellite according to the application scene requirements, and is not specifically limited herein.
And S120, extracting the target elements of the remote sensing image for multiple times to obtain multiple extraction results, and integrating the multiple extraction results to obtain a final target element extraction result.
Before the target elements are segmented from the remote sensing image, the target elements in the remote sensing image to be segmented need to be extracted.
In the disclosure, a plurality of classifiers may be preset, so that a plurality of target element extractions are performed on a remote sensing image through the preset plurality of classifiers to obtain extraction results of the plurality of target elements. For example, T classifiers are preset, and when a remote sensing image is subjected to target extraction for a plurality of times, the remote sensing image is input into the T classifiers, and each classifier outputs an extraction result of one target element for the remote sensing image, so that the extraction results of the T target elements are obtained in total. It should be noted that the classifier reflects a mapping relationship between the remote sensing image and an extraction result of the target element in the remote sensing image.
After the extraction results of the multiple target elements are obtained, the extraction results of the multiple target elements can be integrated according to a preset rule to obtain a final target element extraction result. The preset rule may be a few subject to majority, that is, the extraction result of the class with the largest number is taken as the extraction result of the final target element. It can be understood that even if one or a small number of the extraction results of the plurality of target elements have a deviation, the deviation can be corrected back in the process of obtaining the final target element extraction result by integrating the extraction results of the plurality of target elements, so that the accuracy of the final target element extraction result is improved.
In a possible implementation manner, the remote sensing image is subjected to multiple target element extraction to obtain multiple extraction results, and the final target element extraction result is obtained by integrating the multiple extraction results, and the extraction is performed through a pre-constructed strong classifier. The strong classifier is constructed by adopting an integrated learning algorithm.
The ensemble learning algorithm is an algorithm for obtaining a strong classifier from a plurality of weak classifiers through a certain strategy. The advantages of each weak classifier can be exerted through integrated learning, and a strong classifier with higher generalization capability and robustness is formed. The ensemble learning algorithm may be a Boosting algorithm, a Stacking algorithm, or a Bagging algorithm, which is not specifically limited herein.
In one possible implementation, the strong classifier is constructed by using an ensemble learning algorithm, which includes steps S121-S123.
S121, acquiring a training data set and at least two basic weak classifiers.
The training data set comprises a plurality of sample data, and each sample data comprises a sample remote sensing image and label data corresponding to the sample remote sensing image. And the label data of the sample remote sensing image reflects the correct extraction result of the target elements in the sample remote sensing image.
In one possible implementation manner, the sample data in the training data set is constructed in the following manner, specifically including steps S121-1 to S121-4.
And S121-1, selecting a sample remote sensing image.
When the sample remote sensing image is selected, at least one factor of the coverage area of the remote sensing image, elements included in the remote sensing image and the resolution of the remote sensing image can be considered to select the sample remote sensing image.
For example, the application scenario of the strong classifier is to extract a first setting element from the remote sensing image, and in this case, when the sample remote sensing image is selected, the remote sensing image including the first setting element may be directly selected as the sample remote sensing image. The first setting element may be a building, a vehicle, or other interested analysis elements, and is not limited in particular.
For another example, the application scenario of the strong classifier is to extract the second setting element of the a region from the remote sensing image of the a region, and at this time, when the sample remote sensing image is selected, the remote sensing image including the a region may be selected from the remote sensing image, and then the remote sensing image including the second setting element may be selected from the remote sensing image including the a region as the sample remote sensing image. The second setting element may be a road, a building, a vehicle, or other interested analysis element, and is not limited in particular.
In order to improve the accuracy of the extraction result of the strong classifier, the selected sample remote sensing image also needs to meet a set resolution requirement, for example, the resolution is less than 2m, or the resolution is less than 0.8 m. For example, after the remote sensing image including the second setting element is selected from the remote sensing images including the a area, the remote sensing image with the resolution less than 0.8m can be continuously selected as the sample remote sensing image.
When selecting sample data, the selection of the remote sensing image of the sample can be performed by considering other factors such as background complexity of the remote sensing image, and the like, and is not particularly limited herein.
The quantity of the selected sample data can be set in advance according to the requirements of the application scene.
And S121-2, drawing vector data of the target elements in the same coordinate system with the sample remote sensing image.
The target element is selected according to the segmentation task. For example, if the segmentation task is to segment a building from the remote sensing image, the target element is the building. And if the segmentation task is to segment a road in the remote sensing image, the target element is the road.
When vector data of a target element in a sample remote sensing image is drawn, a coordinate system which is the same as that of the sample remote sensing image is selected, the sample remote sensing image is imported as a base map, and the vector data of the target element in the coordinate system which is the same as that of the sample remote sensing image is drawn in a set map layer based on the sample remote sensing image.
And S121-3, converting the vector data of the target element into raster data with the size consistent with that of the sample remote sensing image, and taking the raster data as label data corresponding to the sample remote sensing image. Since the size of the sample remote sensing image is generally large relative to the raster data after conversion, it is necessary to convert the vector data of the target element into raster data having a size equal to that of the sample remote sensing image.
And S121-4, obtaining sample data according to the sample remote sensing image and the label data. For example, the sample remote sensing image and the label data corresponding to the sample remote sensing image may be used as one sample data.
In a possible implementation manner, when the obtained sample data is less, data enhancement processing such as rotation and noise addition can be performed on the sample data to realize expansion of the sample data.
Before constructing a strong classifier according to an ensemble learning algorithm, at least two basic weak classifiers need to be selected in advance. For example, 2 basic weak classifiers, 5 basic weak classifiers, and 7 basic weak classifiers may be selected according to specific application requirements, where specific data of the basic weak classifiers are not specifically limited.
The at least two selected basic weak classifiers may be of the same type, for example, may both be decision tree weak classifiers, or both be neural network weak classifiers. Different types of weak classifiers are also available, for example, in the case of selecting 2 basic weak classifiers, one weak classifier may be a support vector machine weak classifier, and the other weak classifier may be a logistic regression weak classifier. The type of the at least two basic weak classifiers is not particularly limited herein.
And S122, respectively training the at least two basic weak classifiers based on the training data set to obtain the at least two weak classifiers.
For different ensemble learning algorithms, the method for training at least two basic weak classifiers based on the training data set is also different, and step S122 is described below by taking the Bagging algorithm as an example.
In a possible implementation method, the selected at least two weak classifiers are neural network weak classifiers, and in the Bagging algorithm, the training is performed on at least two basic weak classifiers based on a training data set, including steps S122-1 to S122-2.
S122-1, extracting the training subsets corresponding to at least two basic weak classifiers in the training data set by the extraction method with replacement.
The following describes the extraction process of the training subsets corresponding to each basic weak classifier by taking the extraction of the training subsets corresponding to one basic weak classifier as an example.
For example, the training subset corresponding to the basic weak classifier includes m sample data, and in the process of extracting the training subset, one sample data is randomly extracted from the training data set each time and put into the training subset, and then the sample data is put back, that is, the sample data may still be extracted in the next extraction, so that the training subset including m sample data can be obtained by extracting m times.
The training subsets corresponding to each weak classifier are randomly extracted, so that the training subsets corresponding to each weak classifier are different from the training subsets corresponding to other weak classifiers, and a plurality of different weak classifiers can be obtained based on training of a plurality of different training subsets.
And S122-2, training the basic weak classifiers through the training subsets corresponding to the basic weak classifiers to obtain each weak classifier.
And S123, obtaining the strong classifier for at least two weak classifiers through a preset combination strategy.
And respectively inputting the remote sensing image to be segmented into at least two weak classifiers to obtain the extraction results of a plurality of target elements. For example, the number of the weak classifiers is 2, the remote sensing image to be segmented is respectively input into the two weak classifiers, and the extraction results of 2 target elements are obtained. For another example, the number of the weak classifiers is 9, the remote sensing images to be segmented are respectively input into the 9 weak classifiers, and extraction results of 9 target elements are obtained.
The combination strategy is a strategy for synthesizing a plurality of extraction results to obtain a final target element extraction result. In one possible implementation, the join policy may be a voting policy. The voting strategy can be minority-compliant majority, i.e. the extraction result with the largest number is taken as the final target element extraction result. The combination strategy can also be a combination strategy based on a learning method, namely, a plurality of extraction results are input to a weight learning device, and a final target element extraction result is obtained through the weight learning device. The binding strategy is not specifically limited herein.
In one possible implementation, the weak classifier can be implemented using a U-Net network.
In a high-resolution remote sensing image, spectra corresponding to the same element may be different, namely, the phenomenon of same object and different spectrum exists, so that the shape difference of the same object is large, the same element is difficult to be completely extracted through the existing U-Net network, and the recall ratio of target element extraction is reduced.
In a possible implementation mode, the existing U-Net network is improved, and the improved U-Net network comprises six sequentially cascaded downsampling coding layers and six sequentially cascaded upsampling decoding layers. In the implementation mode, the number of the coding layers and the decoding layers of the U-Net network is expanded, so that the improved U-Net network can extract deeper semantic information from the remote sensing image, the recall ratio of target element extraction is improved, and the accuracy of the final target element extraction result is improved. Therefore, the target elements can be accurately segmented in the remote sensing image based on the final target element extraction result, and the accuracy of target element segmentation is improved.
In the high-resolution remote sensing image, the situation that the same spectrum corresponds to different ground objects exists at the same time, namely, the phenomenon of foreign matters in the same spectrum exists, so that the background of the high-resolution remote sensing image is very complex, the improved U-Net network is difficult to accurately extract target elements from the complex background, and the accuracy of extracting the target elements needs to be further optimized.
In one possible implementation, the improved U-Net network can be further optimized. The structure of the optimized U-Net network is shown in fig. 2, and includes six down-sampling coding layers sequentially cascaded on the left side and six up-sampling decoding layers sequentially cascaded on the right side, and a channel space attention module is cascaded after the last convolution layer of each up-sampling decoding layer.
In the implementation mode, the optimized U-Net network can have higher weight distribution on the extracted target elements by introducing the channel space attention module, so that the interference of a complex background in the remote sensing image can be reduced.
When the U-Net network shown in FIG. 2 is adopted for extracting the target elements, the problems of interference of a complex background of a high-resolution remote sensing image and large morphological difference of the same ground object can be solved, and the accuracy of extracting the target elements in the high-resolution remote sensing image is improved, so that the target elements can be accurately segmented in the remote sensing image based on the final target element extraction result, and the accuracy of segmenting the target elements is improved.
And S130, dividing the target element in the remote sensing image based on the final target element extraction result.
And the final target element extraction result is raster data, the raster data is converted into vector data, and the vector data is loaded to the upper layer of the remote sensing image, so that the target element can be segmented in the remote sensing image.
< method example >
FIG. 3 shows a schematic flow diagram of an example of a method of remote sensing image segmentation according to an embodiment of the present disclosure. As shown in FIG. 3, the remote sensing image segmentation method comprises steps S310-S350.
S310, selecting a sample remote sensing image, and drawing label data corresponding to the sample remote sensing image. The steps of drawing the label data corresponding to the sample remote sensing image refer to steps S121-1 to S121-3, which are not described herein again.
And S320, performing data enhancement processing such as rotation and noise addition on the sample data to obtain expanded sample data.
S330, selecting T U-Net networks as shown in FIG. 2, obtaining T different sample data sets from the extended sample data obtained in the step S320 by a sample data acquisition method with sample replacement, and training the T U-Net networks based on the T sample data sets to obtain T weak classifiers.
S340, obtaining a strong classifier according to the T weak classifiers and the voting strategy.
And S350, inputting the acquired remote sensing image to be segmented into a strong classifier to obtain a final target element extraction result, and segmenting the target element in the remote sensing image based on the final target element extraction result.
According to the remote sensing image segmentation method, a Bagging ensemble learning method is selected according to the characteristic that sample data morphology difference is large, different sample sets are selected to train a plurality of weak classifiers, a final strong classifier is obtained through a voting strategy, the high-resolution remote sensing image segmentation accuracy and recall ratio are improved, and the generalization capability and robustness of high-resolution remote sensing image segmentation are improved. The comparison graph of the remote sensing image segmentation method in the example and the segmentation results of other network structures is shown in fig. 4, and obviously, the remote sensing image segmentation method in the example can remarkably improve the accuracy of image segmentation. Specifically, Source in fig. 4 represents an original remote sensing image, and groudtruth represents a sample true value drawn manually.
< apparatus embodiment >
FIG. 5 shows a schematic block diagram of a remote sensing image segmentation apparatus according to an embodiment of the present disclosure. As shown in fig. 5, the remote sensing image segmentation apparatus 5000 includes:
the image acquisition module 5100 is used for acquiring a remote sensing image to be segmented;
the target extraction module 5200 is configured to perform multiple target element extractions on the remote sensing image to obtain multiple extraction results, and synthesize the multiple extraction results to obtain a final target element extraction result;
and the target segmentation module 5300 is used for segmenting the target element in the remote sensing image based on the final target element extraction result.
In a possible implementation mode, extracting target elements for multiple times from a remote sensing image to obtain multiple extraction results, and synthesizing the multiple extraction results to obtain a final target element extraction result, wherein the extraction is performed through a pre-constructed strong classifier; the strong classifier is constructed by adopting an integrated learning algorithm.
In a possible implementation manner, the remote sensing image segmentation apparatus 5000 further includes a strong classifier construction module, where the strong classifier construction module is specifically configured to obtain a training data set and at least two basic weak classifiers when constructing a strong classifier by using an integrated learning algorithm; respectively training at least two basic weak classifiers based on a training data set to obtain at least two weak classifiers; and for at least two weak classifiers, obtaining a strong classifier through a preset combination strategy.
In one possible implementation, the sample data in the training dataset is constructed by: selecting a sample remote sensing image; drawing vector data of target elements in the same coordinate system with the sample remote sensing image according to the target extraction task; converting the vector data of the target element into raster data with the size consistent with that of the sample remote sensing image, and taking the raster data as label data corresponding to the sample remote sensing image; and obtaining sample data according to the sample remote sensing image and the label data.
In a possible implementation manner, the strong classifier construction module is specifically configured to extract a training subset corresponding to the basic weak classifier in the training data set by a replacement extraction method when training the basic weak classifier based on the training data set; and training the basic weak classifiers through the training subsets corresponding to the basic weak classifiers to obtain each weak classifier.
In one possible implementation, the weak classifier is implemented using a U-Net network.
In one possible implementation manner, the U-Net network comprises a six-layer downsampling coding layer which is sequentially cascaded and a six-layer upsampling decoding layer which is sequentially cascaded.
In one possible implementation, each up-sampling decoding layer includes a channel space attention module; the channel space attention module is cascaded after the last convolution layer of each up-sampling decoding layer.
< electronic device embodiment >
FIG. 6 shows a schematic block diagram of remote sensing image segmentation electronics, in accordance with an embodiment of the present disclosure.
As shown in fig. 6, the remote sensing image segmentation electronic device 6000 includes:
a processor 6100 and a memory 6200 for storing instructions executable by the processor 6100. Wherein the processor 6100 is configured to execute the executable instructions to implement the remote sensing image segmentation method of any of the above.
Here, it should be noted that the number of the processors 6100 may be one or more. Meanwhile, in the remote sensing image segmentation electronic device 6000 of the embodiment of the present disclosure, an input device 6300 and an output device 6400 may also be included. The processor 6100, the memory 6200, the input device 6300, and the output device 6400 may be connected by a bus, or may be connected by another method, and the connection is not limited specifically here.
The memory 6200, which is a computer-readable storage medium, may be used to store software programs, computer-executable programs, and various modules, such as: the remote sensing image segmentation method provided by the embodiment of the disclosure corresponds to a program or a module. The processor 6100 executes various functional applications and data processing of the remote sensing image segmentation electronic device 6000 by running software programs or modules stored in the memory 6200.
The input device 6300 may be used to receive input numbers or signals. Wherein the signal may be a key signal generated in connection with user settings and function control of the device/terminal/server. The output device 6400 may include a display device such as a display screen.
Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terms used herein were chosen in order to best explain the principles of the embodiments, the practical application, or technical improvements to the techniques in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.
Claims (10)
1. A remote sensing image segmentation method is characterized by comprising the following steps:
acquiring a remote sensing image to be segmented;
extracting target elements for multiple times from the remote sensing image to obtain multiple extraction results, and synthesizing the multiple extraction results to obtain a final target element extraction result;
and based on the final target element extraction result, dividing the target element in the remote sensing image.
2. The method according to claim 1, characterized in that, the remote sensing image is subjected to a plurality of times of target element extraction to obtain a plurality of extraction results, and the final target element extraction result is obtained by integrating a plurality of extraction results, and the extraction is carried out through a pre-constructed strong classifier;
and the strong classifier is constructed by adopting an integrated learning algorithm.
3. The method of claim 2, wherein constructing the strong classifier using an ensemble learning algorithm comprises:
acquiring a training data set and at least two basic weak classifiers;
respectively training at least two basic weak classifiers based on the training data set to obtain at least two weak classifiers;
and for at least two weak classifiers, obtaining the strong classifier through a preset combination strategy.
4. The method of claim 3, wherein the sample data in the training data set is constructed by:
selecting a sample remote sensing image;
drawing vector data of target elements in the same coordinate system with the sample remote sensing image according to a target extraction task;
converting the vector data of the target element into raster data with the size consistent with that of the sample remote sensing image, and taking the raster data as label data corresponding to the sample remote sensing image;
and obtaining the sample data according to the sample remote sensing image and the label data.
5. The method of claim 3, wherein training the at least two basic weak classifiers based on the training data set comprises:
extracting a training subset corresponding to the basic weak classifier from the training data set by a replaced extraction method;
and training the basic weak classifiers through the training subsets corresponding to the basic weak classifiers to obtain the weak classifiers.
6. The method of claim 3, wherein the weak classifier is implemented using a U-Net network.
7. The method of claim 6, wherein the U-Net network comprises a sequentially cascaded six-layer down-sampling coding layer and a sequentially cascaded six-layer up-sampling decoding layer.
8. The method of claim 7, wherein each of said up-sampling decoding layers comprises a channel space attention module;
the channel space attention module is cascaded after the last convolution layer of each up-sampling decoding layer.
9. A remote sensing image segmentation apparatus, comprising:
the image acquisition module is used for acquiring a remote sensing image to be segmented;
the target extraction module is used for extracting target elements from the remote sensing image for multiple times to obtain multiple extraction results, and synthesizing the multiple extraction results to obtain a final target element extraction result;
and the target segmentation module is used for segmenting the target element in the remote sensing image based on the final target element extraction result.
10. An electronic device for remote sensing image segmentation, comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to implement the method of any one of claims 1 to 8 when executing the executable instructions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111665954.0A CN114283286A (en) | 2021-12-30 | 2021-12-30 | Remote sensing image segmentation method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111665954.0A CN114283286A (en) | 2021-12-30 | 2021-12-30 | Remote sensing image segmentation method and device and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114283286A true CN114283286A (en) | 2022-04-05 |
Family
ID=80879299
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111665954.0A Pending CN114283286A (en) | 2021-12-30 | 2021-12-30 | Remote sensing image segmentation method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114283286A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945374A (en) * | 2012-10-24 | 2013-02-27 | 北京航空航天大学 | Method for automatically detecting civil aircraft in high-resolution remote sensing image |
| KR20150103447A (en) * | 2014-03-03 | 2015-09-11 | 에스케이플래닛 주식회사 | Apparatus for classifying an image, method thereof and computer readable medium having computer program recorded therefor |
| CN109034233A (en) * | 2018-07-18 | 2018-12-18 | 武汉大学 | A kind of high-resolution remote sensing image multi classifier combination classification method of combination OpenStreetMap |
| CN111914611A (en) * | 2020-05-09 | 2020-11-10 | 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 |
| CN113780174A (en) * | 2021-09-10 | 2021-12-10 | 福州大学 | High vegetation platform storm and rain type landslide identification method combined with random forest algorithm |
-
2021
- 2021-12-30 CN CN202111665954.0A patent/CN114283286A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945374A (en) * | 2012-10-24 | 2013-02-27 | 北京航空航天大学 | Method for automatically detecting civil aircraft in high-resolution remote sensing image |
| KR20150103447A (en) * | 2014-03-03 | 2015-09-11 | 에스케이플래닛 주식회사 | Apparatus for classifying an image, method thereof and computer readable medium having computer program recorded therefor |
| CN109034233A (en) * | 2018-07-18 | 2018-12-18 | 武汉大学 | A kind of high-resolution remote sensing image multi classifier combination classification method of combination OpenStreetMap |
| CN111914611A (en) * | 2020-05-09 | 2020-11-10 | 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 |
| CN113780174A (en) * | 2021-09-10 | 2021-12-10 | 福州大学 | High vegetation platform storm and rain type landslide identification method combined with random forest algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bielski et al. | Emergence of object segmentation in perturbed generative models | |
| US20210183022A1 (en) | Image inpainting method and apparatus, computer device, and storage medium | |
| CN111047516B (en) | Image processing method, image processing device, computer equipment and storage medium | |
| US10970543B2 (en) | Distributed and self-validating computer vision for dense object detection in digital images | |
| US8160366B2 (en) | Object recognition device, object recognition method, program for object recognition method, and recording medium having recorded thereon program for object recognition method | |
| CN111369581A (en) | Image processing method, device, equipment and storage medium | |
| CN108961180B (en) | Infrared image enhancement method and system | |
| Vaishnavi et al. | Application of local invariant symmetry features to detect and localize image copy move forgeries | |
| CN113901900A (en) | Unsupervised change detection method and system for homologous or heterologous remote sensing image | |
| CN104094284A (en) | Image processor with edge-preserving noise suppression functionality | |
| CN110211046B (en) | Remote sensing image fusion method, system and terminal based on generation countermeasure network | |
| CN107688783B (en) | 3D image detection method and device, electronic equipment and computer readable medium | |
| CN113496150B (en) | Dense target detection method and device, storage medium and computer equipment | |
| CN109977832B (en) | Image processing method, device and storage medium | |
| Herrault et al. | Automatic extraction of forests from historical maps based on unsupervised classification in the CIELab color space | |
| Attarzadeh et al. | Object-based rule sets and its transferability for building extraction from high resolution satellite imagery | |
| CN113901972A (en) | Method, device and equipment for detecting remote sensing image building and storage medium | |
| CN113378897A (en) | Neural network-based remote sensing image classification method, computing device and storage medium | |
| CN115661611A (en) | Infrared small target detection method based on improved Yolov5 network | |
| Serwa et al. | Enhancement of classification accuracy of multi-spectral satellites’ images using Laplacian pyramids | |
| Hu et al. | Classification of very high-resolution remote sensing imagery using a fully convolutional network with global and local context information enhancements | |
| CN114998630B (en) | Ground-to-air image registration method from coarse to fine | |
| US10210621B2 (en) | Normalized probability of change algorithm for image processing | |
| CN116310832A (en) | Remote sensing image processing method, device, equipment, medium and product | |
| CN114283286A (en) | Remote sensing image segmentation method and device and electronic equipment |
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 |