CN112053371A

CN112053371A - Water body extraction method and device in remote sensing image

Info

Publication number: CN112053371A
Application number: CN202011051057.6A
Authority: CN
Inventors: 陆川; 周舒婷; 罗钦瀚; 徐康
Original assignee: Chengdu Star Age Aerospace Technology Co ltd
Current assignee: Chengdu Star Age Aerospace Technology Co ltd
Priority date: 2020-09-29
Filing date: 2020-09-29
Publication date: 2020-12-08

Abstract

The utility model discloses a method for extracting water in remote sensing images, which comprises the steps of obtaining the remote sensing images to be extracted; performing superpixel segmentation on the obtained remote sensing image to obtain a plurality of superpixel regions, and determining the statistical characteristics of each superpixel region; determining the region attribute of each super-pixel region by utilizing a pre-trained water body extraction model for classification according to the statistical characteristics of each super-pixel region; and marking the area corresponding to the remote sensing image to be subjected to water body extraction as the water body according to the super pixel area determined as the water body, thereby obtaining a final water body extraction result graph. A determination method of the water body extraction model in the remote sensing image is also disclosed.

Description

Water body extraction method and device in remote sensing image

Technical Field

The present disclosure relates to, but not limited to, the technical field of remote sensing image processing, and in particular, to a method and an apparatus for extracting a water body from a remote sensing image.

Background

The water body is one of important basic geographic information, is the most common ground object type in the remote sensing image, and the rapid acquisition of the dynamic information of the water body has very obvious practical value and scientific significance to the career of water resource investigation, water conservancy planning, environmental monitoring, protection and the like.

At present, there are many algorithmic studies aiming at water body extraction in remote sensing images, but there are the following two problems:

firstly, most of the current mainstream algorithms utilize spectral information in a remote sensing image to calculate a single pixel (such as an NDVI index) in the image, and judge that the pixel belongs to land or water through a preset threshold, such methods do not fully utilize image information quantities such as geometric shapes of remote sensing ground objects, adjacent pixel correlation and the like, and the extracted water result is prone to have a 'salt and pepper phenomenon', is not easy to vectorize and has unclear boundaries.

Secondly, because most water bodies are affected differently by silt, suspended matters, freezing and depth, and factors such as illumination and cloud cover shielding during remote sensing image imaging, the imaging difference of surface water body data in the same geographic range on different remote sensing images is large, only pixel information in the images is utilized, and the extraction effect of the actual water body in a preset threshold value or a classifier model is often not accurate.

Disclosure of Invention

The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

The embodiment of the disclosure provides a method or a device for extracting a water body in a remote sensing image, which can remarkably improve the accuracy of extracting a water body region in the remote sensing image.

The embodiment of the present disclosure provides a method for extracting a water body from a remote sensing image, which includes,

acquiring a remote sensing image to be subjected to water body extraction;

performing superpixel segmentation on the obtained remote sensing image to obtain a plurality of superpixel regions, and determining the statistical characteristics of each superpixel region;

determining the region attribute of each super-pixel region by utilizing a pre-trained water body extraction model for classification according to the statistical characteristics of each super-pixel region;

and marking the area corresponding to the remote sensing image to be subjected to water body extraction as the water body according to the super pixel area determined as the water body, thereby obtaining a final water body extraction result graph.

In some exemplary embodiments, the water body extraction model for classification is trained in advance according to the following method:

obtaining a remote sensing image as a sample;

acquiring water body vector data of a corresponding geographic area in a pre-established water body vector data set according to the acquired remote sensing image;

selecting area attributes corresponding to a preset number of super-pixel area marks according to the water body vector data, wherein the area attributes comprise: water body, land;

and training a Support Vector Machine (SVM) classifier serving as the water body extraction model according to the region attributes and the statistical characteristics of the super-pixel regions with the preset number.

In some exemplary embodiments, the pre-established set of water vector data comprises: and a water body vector database which is established in advance according to the geographic information provided by the third-party geographic information system.

In some exemplary embodiments, the obtaining, from the obtained remote sensing image, water vector data of a corresponding geographic area in a pre-established water vector data set includes:

acquiring at least one target water body vector data in a geographical range included by the remote sensing image from a pre-established water body vector data set according to geographical range information included by the acquired remote sensing image;

constructing and determining a target water body vector diagram according to the at least one target water body vector data;

and converting the target water body vector diagram into a binaryzation water body image.

In some exemplary embodiments, the performing the superpixel segmentation on the obtained remote sensing image to obtain a plurality of superpixel regions includes:

and segmenting the acquired remote sensing image by using a simple linear iterative clustering method to obtain the plurality of super pixel regions.

In some exemplary embodiments, the training of the SVM classifier as the water body extraction model according to the region attributes and the statistical features of the predetermined number of superpixel regions includes:

taking the super-pixel areas marked as water bodies in the super-pixel areas with the preset number as positive samples, and taking the super-pixel areas marked as land in the super-pixel areas with the preset number as negative samples;

and training the SVM classifier according to the statistical characteristics of the positive samples and the statistical characteristics of the negative samples.

In some exemplary embodiments, the determining, according to the statistical characteristics of each super-pixel region, the region attribute of each super-pixel region by using a water body extraction model trained in advance for classification includes:

the following steps are respectively performed for each super pixel region:

determining the sub-weight of each dimension of statistical characteristics of the super-pixel region by utilizing a pre-trained water body extraction model for classification,

determining the comprehensive weight of the super-pixel region according to the sub-weights of all the statistical characteristics;

determining the region attribute of the super pixel region according to the comprehensive weight, the first weight standard and the second weight standard; the area attribute corresponding to the first weight standard is a water body, and the area attribute corresponding to the second weight standard is a land.

In some exemplary embodiments, the statistical features include: the method comprises the steps of full variation characteristics, NIR channel histogram characteristics of a near infrared spectrum of a super pixel region and histogram characteristics of a Gabor filtered channel of the super pixel region.

The embodiment of the disclosure further provides an electronic device, which includes a memory and a processor, where the memory stores a computer program for performing water body extraction in a remote sensing image, and the processor is configured to read and run the computer program for performing water body extraction in the remote sensing image to execute the above-mentioned water body extraction method in the remote sensing image.

The embodiment of the disclosure also provides a storage medium, in which a computer program is stored, where the computer program is configured to execute the above-mentioned water body extraction method in a remote sensing image when running.

Other aspects will be apparent upon reading and understanding the attached drawings and detailed description.

Drawings

FIG. 1 is a flowchart of a method for extracting a water body from a remote sensing image according to an embodiment of the present disclosure;

FIG. 2 is a flow chart of a method for determining a water body extraction model in a remote sensing image according to another embodiment of the present disclosure;

FIG. 3 is a schematic diagram of an initial seed point in an embodiment of the present disclosure;

FIG. 4 is a schematic diagram of a seed point after the position has been moved in an embodiment of the present disclosure;

FIG. 5 is a flowchart of a method for extracting water from a remote sensing image according to another embodiment of the present disclosure;

FIG. 6 is a flowchart of a method for determining a water body extraction model in a remote sensing image according to another embodiment of the present disclosure;

fig. 7 is a flowchart of a water body extraction method in a remote sensing image according to another embodiment of the disclosure.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments and features of the embodiments in the present application may be arbitrarily combined with each other without conflict.

The following step numbers do not limit a specific execution order, and the execution order of some steps can be adjusted according to specific embodiments.

Example one

The embodiment of the present disclosure provides a method for extracting a water body from a remote sensing image, as shown in fig. 1, including,

step 11, acquiring water body vector data from geographic information of a third-party data source, and establishing a water body vector database, wherein the water body vector database is used for follow-up query; the third-party data source comprises: *** maps, Gade maps, etc. Not limited to geographic information data of a particular company. Acquiring water body vector data from geographic information of a third-party data source according to a related technical method by a person skilled in the art, and establishing a water body vector database; the method for acquiring water vector data in the embodiments of the present disclosure is not limited to a specific manner, and the specific scheme does not belong to the scope defined in the present application.

Step 12, obtaining a remote sensing image as a sample;

step 13, acquiring corresponding water vector data of the geographical area according to the remote sensing image;

step 14, performing superpixel segmentation on the remote sensing image serving as the sample to obtain a plurality of superpixel regions;

step 15, calculating the statistical characteristics of each super-pixel region, namely calculating the statistical characteristics of each super-pixel region according to the acquired water body vector data of the geographic region and the plurality of super-pixel regions of the remote sensing image serving as the sample;

step 16, selecting a preset number of super-pixel areas in the remote sensing image of the sample as sample super-pixel areas, and marking the area attribute of each sample super-pixel area; the region attributes include at least: water body and land. And taking the marked super-pixel area as a training sample.

Step 17, training an SVM classifier; namely, training an SVM (Support Vector Machine) classifier according to the statistical characteristics of the training samples; the trained SVM classifier is an extraction model for subsequently extracting the water body, and the SVM comprises relevant weights.

Step 22, obtaining a remote sensing image to be subjected to water body extraction;

step 23, acquiring corresponding water vector data of the geographical area according to the remote sensing image;

24, performing superpixel segmentation on the remote sensing image of the water body to be extracted to obtain a plurality of superpixel regions;

step 25, calculating the statistical characteristics of each super-pixel region according to the geographical region water vector data corresponding to the remote sensing image;

step 26, SVM judgment; namely, determining the region attribute of each superpixel region by using the SVM classifier trained in the step 17;

step 27, determining a water body extraction result; namely, the super-pixel region of the water body is determined, and the region corresponding to the remote sensing image to be subjected to water body extraction is marked as the water body, so that a final water body extraction result graph is obtained.

In some exemplary embodiments, step 11 is a preparation step performed in advance, and may be performed in advance, and is not limited to being performed with a subsequent classifier training or water body extraction process.

In some exemplary embodiments, steps 12-17 are a classifier training process performed by using samples, and the trained classifier may be used for water extraction in multiple remote sensing images, and the training process is not limited to be performed for each water extraction. In some exemplary embodiments, steps 12-17 are accomplished in advance.

In some exemplary embodiments, the SVM classifier in step 17 is a linear SVM classifier or a non-linear SVM classifier.

In some exemplary embodiments, steps 22-27 are a process of performing water body extraction on a remote sensing image to be subjected to water body extraction. The steps 23 to 25 are identical to the steps 13 to 15 in processing manner, but differ in processing object, and the step 13 to 15 is a step performed on the remote sensing image as a sample.

In some exemplary embodiments, steps 12-17 include the following steps, as shown in FIG. 2:

step 12 comprises: step 121, obtaining a remote sensing image as a sample;

step 13 comprises: step 131-;

step 131, determining a geographical range corresponding to the remote sensing image;

step 132, acquiring target water body vector data in a geographic range;

step 133, converting the target water body vector data into a binary image (binary water body image), and taking the binary image as geographical area water body vector data corresponding to the remote sensing image of the sample;

step 14 comprises: step 141, performing superpixel segmentation on the remote sensing image serving as the sample by using a Simple Linear Iterative Clustering (SLIC) method to obtain a plurality of superpixel regions;

step 15 comprises: step 151, calculating the statistical characteristics of each super pixel region; wherein the statistical features at least include: full variation characteristics, NIR channel histogram characteristics, histogram characteristics after Gabor filtration and water body characteristics;

step 16 comprises: step 161, selecting and marking training samples;

step 17 comprises: step 171, training the SVM classifier.

In some exemplary embodiments, the step 23 includes the step 131 and the step 133, the step 24 includes the step 141, and the step 25 includes the step 151, wherein the processed remote sensing image is replaced by a remote sensing image of the water body to be extracted.

In some exemplary embodiments, the step 131 of determining the corresponding geographic range of the remote sensing image comprises: and acquiring geographical range information included in the remote sensing image by using a GDAL (geographic Data Abstraction Library), namely a longitude range interval and a latitude range interval corresponding to the remote sensing image.

In some exemplary embodiments, step 132 obtains target water vector data over a geographic area, including:

acquiring target water body vector data corresponding to the geographical range included by the remote sensing image from the established water body vector database according to the geographical range information included by the remote sensing image; in some exemplary embodiments, the water body regions in the water body vector database are independent from each other, and therefore in this step, the corresponding operation is to use the geographical range included in the remote sensing image and the corresponding water body regions to find the intersection, and use the obtained water body region of the intersection as the target water body vector data. Obtaining a plurality of target water body vector data by the method; and splicing the obtained multiple water body vector data into a target water body vector diagram according to the geographic space information of the multiple water body vector data, wherein the target water body vector diagram is the same as the remote sensing image in size. The water body vector database comprises vector data of water body areas in various geographic spaces. Therefore, intersection is obtained by utilizing the geographical range included by the remote sensing image and corresponding water body areas, and the intersection is obtained according to the geographical range included by the remote sensing image and the vector data of the water body areas, so that the intersection part of the water body areas and the remote sensing image can be obtained.

acquiring water body vector data in the geographical range included by the remote sensing image from a water body vector database according to the geographical range information included by the remote sensing image; and splicing the obtained multiple water body vector data into a target water body vector diagram according to the geographic space information of the multiple water body vector data.

In some exemplary embodiments, step 133 converts the target water vector data into a binary image, including:

the target water body vector image is converted into a binary image, also called a binary water body image or a binary water body image (the image size is consistent with the remote sensing image), by using the GDAL, wherein the pixel value of the pixels of the water body region in the binary water body image is set to 255, and the pixel value of the pixels of other regions is set to 0.

In some exemplary embodiments, the segmentation of the remotely sensed image into a plurality of super-pixel regions is performed, in particular, using SLIC (simple linear iterative clustering). The SLIC is a segmentation method which effectively generates compact almost unified superpixels by clustering pixels in a remote sensing image by using the color similarity of the pixels and an image one-plane space. Wherein step 14 or 141 comprises:

1. obtaining remote sensing images of the sample, and carrying out channel separation on the images according to R, G, B, NIR to respectively obtain images of R, G, B, NIR channels;

2. fusing the R, G, B channel images obtained by separation to obtain a natural color first image, and converting the natural color first image into a Lab color space first image;

3. setting a preset number of initialization seed points (cluster centers) in a first image converted into a Lab color space, specifically comprising: according to the preset number of super pixels, uniformly distributing seed points according to the preset number of super pixels in the first image converted into the Lab color space, wherein the number of the seed points is the same as the preset number of the super pixels; (the black pixel point is set as the seed point position as shown in FIG. 3)

4. Reselecting the position of the seed point in the 3 x 3 neighborhood of the seed point, comprising: calculating gradient values of all pixel points in a 3 x 3 neighborhood of the seed point, and moving the seed point to the pixel point position with the minimum gradient value in the neighborhood; (as shown in FIG. 4, shading indicates the position of the transformed seed point)

In some exemplary embodiments, the seed point is selected in a neighborhood of a, where a is an integer greater than 1, and is not limited to a-3 in this embodiment.

5. Assign the label for every pixel in the neighborhood around every seed point after the replacement position, include: traversing all pixel points except the seed points, and calculating the distance measurement between each pixel point and each seed point; and distributing corresponding labels for corresponding pixel points according to the distance measurement between the pixel points and each seed point. Specifically, a certain pixel point and the seed point with the minimum distance measurement are used as the label of the pixel point, and if the distance measurement between the pixel point a and the seed points 1, 2, 3, and 4 is calculated, and the seed point with the minimum distance measurement with the pixel point is obtained as 1, the label of the pixel point is the seed point 1; if the seed point with the minimum distance measurement with the pixel point is 3, the label of the pixel point is the seed point 3. In some exemplary embodiments, the label of a certain pixel point may correspond to a plurality of seed points.

Wherein, step 5 includes determining the label of each pixel point through multiple iterations, including the following steps:

step 51, calculating the distance measurement of each pixel point, and calculating the color distance and the space distance between each pixel in the first image converted into the Lab color space and the adjacent seed point as follows:

d_cis the color distance, d_sIs the spatial distance, D' is a distance measure, N_cAt a maximum color distance, N_sIs the sampling interval. Converting the image from RGB color space to Lab color space, wherein the (l, a, b) color value and (x, y) coordinate corresponding to each pixel form a 5-dimensional vector V [ l, a, b, x, y ]]The similarity of two pixels can be measured by their vector distance D', the greater the distance, the less similar. j, i represent pixel points and seed points, respectively.

And step 52, determining the distribution label of each pixel according to the distance measurement from each seed to each pixel point, wherein the corresponding seed point label is the distribution label of each pixel. Specifically, a certain pixel point and a seed point with the minimum distance measurement are used as labels of the pixel point, for example, the label in step 5, seed 1, seed 3, and the like;

step 53, adjusting (moving) the position of the seed point in the first image of the Lab color space according to the label of each pixel, and recalculating the distance measurement from each pixel point to the seed point after the position is moved according to the position of the seed point after the seed point is moved; repeatedly executing steps 51, 52 and 53 according to a preset number of times P; finally determining the positions of the label and the seed point of each pixel until the last execution is finished; in some exemplary embodiments, P is 10, or other integer greater than 1.

Wherein adjusting (moving) the position of the seed point in the first image of the Lab color space according to the label of each pixel comprises:

and according to the label information of each pixel, forming the pixels with the same label information into a region, calculating the geometric center pixel of each region, and taking the position of the geometric center pixel of each region as the adjusted position of the seed point of the corresponding region.

6. And according to the label of each pixel point, dividing the first image of the Lab color space into a plurality of super pixel areas. Specifically, the pixel points with the same label are divided into the same super pixel region.

7. And (3) enhancing connectivity, redistributing discontinuous super-pixel regions or super-pixel regions with undersize to adjacent super-pixel regions, and adjusting a plurality of super-pixel regions of the first image of the Lab color space.

Wherein the discontinuous super pixel region includes: if a certain super pixel region comprises a super pixel region composed of several other labels, the discontinuous super pixel region is redistributed to the adjacent super pixel region, and the method comprises the following steps: and the super pixel area consisting of a plurality of other labels in the super pixel area is divided into the super pixel area again.

The super pixel region that is too small in size includes: if the pixel size of a certain super pixel region is smaller than a preset value, the super pixel region with the smaller size is redistributed to the adjacent super pixel regions, and the method comprises the following steps: recalculating the distance between the pixels of the super pixel region and the seed points in the adjacent super pixel region, allocating labels to the pixels in the super pixel region according to the seed points with the minimum distance, and dividing the pixels in the super pixel region into the adjacent super pixel region according to the allocated labels. Thereby realizing adjustment of multiple super pixel areas of the first image of the Lab color space

8. And determining the position of each super-pixel region in the remote sensing image according to the adjusted range of each super-pixel region in the first image of the Lab color space. The position of each super-pixel region in the remote sensing image can be determined according to the adjusted position of each super-pixel region in the first image of the Lab color space.

In some exemplary embodiments, step 15 or 151 includes:

1. calculating the total variation characteristics of each super pixel region, comprising: the fully variant feature for each superpixel region in the remote sensing image can be represented by the following formula:

wherein TV (u) represents the total variation characteristic value of a super pixel region, d Ω represents the area of the super pixel region,

the gradient value of a certain pixel point in the super-pixel region is represented, and the gradient of the pixel point in the embodiment of the disclosure is the sum of first-order gradients in the horizontal direction and the vertical direction.

Wherein the content of the first and second substances,

calculated using the following formula:

in the formula, (x, y) represents the position of a pixel point in a remote sensing image, u (x, y) represents the gray value of the pixel point (x, y), u (x-1, y) represents the gray value of the pixel point (x-1, y), | u (x, y) -u (x-1, y) | represents the horizontal gradient at the pixel point (x, y), u (x, y-1) represents the gray value of the pixel point (x, y-1), and | u (x, y) -u (x, y-1) | represents the vertical gradient at the pixel point (x, y). That is, the total variation value (total variation characteristic) of the super-pixel region is the sum of the horizontal gradient value and the vertical gradient value of all the pixels in the region.

2. Determining NIR channel histogram features for each superpixel region; namely, the distribution of NIR spectral values in each super pixel region is counted and normalized. The method comprises the following steps:

a 16-dimensional (16 in the first preset dimension) histogram is selected, and then the NIR value of each super-pixel region is counted to obtain the 16-dimensional features of the super-pixel region. In some exemplary embodiments, the histogram dimension (the first preset dimension) may be other preset values a (a is greater than 0), resulting in corresponding a-dimensional features.

Normalization: and dividing the obtained 16-dimensional feature by the total number of the pixel points of the whole area of the super pixel area to obtain the NIR channel histogram feature of the super pixel area. In some exemplary embodiments, when the histogram dimension is the other preset value a, the obtained a-dimensional feature is divided by the total number of the pixel points of the entire region of the super-pixel region accordingly.

3. Histogram features of the Gabor filtered channels for each superpixel region are determined. Calculating a channel value of each super-pixel region after Gabor filtering, then counting histograms after filtering in different directions, and finally carrying out normalization. The method comprises the following steps:

performing channel separation on each super pixel region according to R, G, B, NIR channels to obtain R, G, B, NIR channel images of each super pixel region;

performing filtering operations on R, G, B channel images of each super pixel region by using a preset number B (B is greater than 0, and the preset number B is 6 in this embodiment) of Gabor filters in different directions, so as to obtain 3 × 6 filtered images of the super pixel region;

and counting the histogram distribution of the image of each filtered super pixel region to obtain 3-dimensional 6-dimensional 16-dimensional features, wherein the 3-dimensional 6-dimensional 16-dimensional features are histogram features of Gabor filtered channels of the corresponding super pixel region. Wherein 16 is the first predetermined dimension in step 2; in some exemplary embodiments, features are obtained in dimensions 3 × B × a.

Normalization: dividing the obtained features by the total number of pixel points in the whole region of the super pixel region, and taking the normalized value as the histogram feature of the Gabor filtered channel of the super pixel region.

4. Determining water body characteristics, including:

determining the water body range in each super pixel region according to the binarized water body image obtained in the step 133 and the pixel position corresponding relation between the binarized water body image and the remote sensing image, calculating the number of pixels of which the pixels are water bodies in each super pixel region according to the water body range in each super pixel region, and setting the water body as water body characteristic 1 if the pixel data exceeds the total number of all pixels included in the corresponding super pixel region by more than a preset proportion (for example, 50 percent), otherwise, setting the water body characteristic as water body characteristic 0; 1 denotes a water body and 0 denotes a land.

5. Determining a final dimension for each super-pixel region, comprising: and calculating the N dimension of each super pixel region, and superposing the dimensions, wherein N is 1+16+3 + 6+ 16+1 is 306 dimensions.

In this embodiment, the total variation feature is a 1-dimensional feature, the NIR channel histogram feature is a 16-dimensional feature, and the histogram feature of the Gabor-filtered channel is a 3 × 6 × 16-dimensional feature; the water body features are 1-dimensional features.

It can be seen that the N-dimensional statistical features of each superpixel region are calculated through steps 1-5 above.

In some exemplary embodiments, step 16 or 161 includes:

selecting a preset number of super-pixel regions from the plurality of super-pixel regions obtained in step 14 or 141 as sample super-pixel regions, and manually marking region attributes for the sample super-pixel regions according to actual conditions, wherein the attribute labels at least comprise: water body, land;

and taking the sample super-pixel area marked as the water body as a positive sample, and taking the sample super-pixel area marked as the land as a negative sample.

In some exemplary embodiments, step 17 or 171 includes:

and training the SVM classifier according to the statistical characteristics of the positive samples and the statistical characteristics of the negative samples, and determining the weight of the SVM classifier.

In some exemplary embodiments, the weights of the SVM classifier include: the sub-weight and the comprehensive weight of each one-dimensional feature;

the training the SVM classifier according to the statistical characteristics of the positive samples and the statistical characteristics of the negative samples to determine the weight of the SVM classifier comprises the following steps:

training the SVM classifier according to the statistical characteristics of the positive samples and the statistical characteristics of the negative samples, and determining the sub-weight of each one-dimensional statistical characteristic;

determining the comprehensive weight of each sample according to the sub-weights of all the statistical characteristics of each sample;

a first weighting criterion is determined based on the integrated weights of all positive samples, and a second weighting criterion is determined based on the integrated weights of all negative samples.

Determining the region attribute of the super-pixel region when the first weight standard and the second weight standard are used for subsequent water body extraction; when the water body is extracted subsequently, after the comprehensive weight of a certain superpixel region is determined by using the trained SVM classifier, when the comprehensive weight meets a first weight standard, the region attribute of the superpixel region is determined to be the water body, and when the comprehensive weight meets a second weight standard, the region attribute of the superpixel region is determined to be the land.

It should be noted that, the manually marking the region attribute for the sample super-pixel region according to the actual situation includes: and manually analyzing the image of the sample superpixel area, including color, so as to determine the area attribute of the sample superpixel area, and setting the area attribute of the sample superpixel area belonging to the water body as the water body, otherwise, setting the area attribute as land.

The water body extraction scheme provided by the disclosure is a remote sensing image water body extraction method based on superpixel segmentation and multisource data, and due to the difference of the update period of the water body vector data, the water body vector data is different from the actual geographic water body of the current time node, and the remote sensing image can be obtained by shooting through a remote sensing satellite in time as required, so that the water body area in the remote sensing image can better accord with the actual geographic water body coverage condition of the current time node. Although the water body area in the remote sensing image can be directly extracted, most of the water body area is extracted singly, and time cost and labor cost are increased if batch operation is needed. And this application has ensured the accuracy and the reliability of water extraction model through having utilized the multisource data characteristic that the water vector data that comes from the third party data source and the remote sensing image combine together, has utilized super pixel to cut apart the accuracy nature of having guaranteed the segmentation edge, and through both combinations, the overall effect of water extraction is better, and the degree of accuracy promotes greatly.

Example two

The embodiment of the present disclosure further provides a method for determining a water body extraction model in a remote sensing image, as shown in fig. 1, including:

step 12, obtaining a remote sensing image as a sample;

step 16, selecting and marking a training sample;

step 17, training an SVM classifier; and the trained SVM classifier is the determined water body extraction model.

In some exemplary embodiments, further implementation steps of the above steps are similar to corresponding steps in the first embodiment, and are not repeated.

EXAMPLE III

The embodiment of the present disclosure further provides a method for extracting a water body from a remote sensing image, as shown in fig. 1, including:

step 25, calculating the statistical characteristics of each super-pixel region, namely calculating the statistical characteristics of each super-pixel region according to the acquired water body vector data of the geographic region and a plurality of super-pixel regions of the remote sensing image of the water body to be extracted;

step 26, SVM judgment; determining the region attribute of each super-pixel region by using the SVM classifier trained in the first or second embodiment;

and 27, determining a water body extraction result.

In some exemplary embodiments, step 26 comprises:

determining the comprehensive weight of each super-pixel region by using a pre-trained water body extraction model for classification; determining a region attribute of each superpixel region according to the integrated weight, the first weight criterion, and the second weight criterion.

In some exemplary embodiments, step 26 comprises:

for each super-pixel region, determining its region attribute by performing the following steps:

determining the sub-weights corresponding to the statistical characteristics of the super-pixel region by using a pre-trained water body extraction model for classification;

determining the comprehensive weight of the super pixel region according to all the sub-weights;

and when the comprehensive weight meets the first weight standard, determining that the regional attribute of the superpixel region is water, and when the comprehensive weight meets the second weight standard, determining that the regional attribute of the superpixel region is land.

In some exemplary embodiments, the water body extraction model trained in advance for classification may be the SVM classifier trained in step 17 in embodiment one or two. In some exemplary embodiments, further implementation steps of the above steps are similar to corresponding steps in the first embodiment, and are not repeated.

Example four

The embodiment of the present disclosure further provides a method for extracting a water body from a remote sensing image, as shown in fig. 5, including:

step 51, obtaining a remote sensing image to be subjected to water body extraction;

step 52, performing superpixel segmentation on the obtained remote sensing image to obtain a plurality of superpixel regions, and determining the statistical characteristics of each superpixel region;

step 53, determining the region attribute of each super-pixel region by using a pre-trained water body extraction model for classification according to the statistical characteristics of each super-pixel region;

and step 54, marking the area corresponding to the remote sensing image to be subjected to water body extraction as the water body according to the super pixel area determined as the water body, thereby obtaining a final water body extraction result graph.

obtaining a remote sensing image as a sample;

constructing a target water body vector diagram according to the at least one target water body vector data;

the following steps are respectively performed for each super pixel region:

In some exemplary embodiments, the segmenting the obtained remote sensing image by using a simple linear iterative clustering method to obtain the plurality of super pixel regions includes:

determining a first image of a corresponding Lab color space according to the obtained remote sensing image;

determining the position of each seed point in the first image of the Lab color space according to the first image of the Lab color space and a preset number of seed points;

determining a seed point label of each pixel point in the first image of the Lab color space according to the position of each seed point in the first image of the Lab color space;

and determining a plurality of super pixel regions corresponding to the obtained remote sensing image according to the seed point label of each pixel point in the first image of the Lab color space.

In some exemplary embodiments, the statistical features include: features of N dimensions;

wherein the N dimensional features include: 1-dimensional full variation characteristic of the super pixel region, M-dimensional characteristic corresponding to near infrared spectrum (NIR) channel histogram of the super pixel region and L-dimensional characteristic corresponding to the Gabor filtered channel histogram of the super pixel region; where N is 1+ M + L, M, L is an integer greater than 0.

In some exemplary embodiments, the 1-dimensional fully-variant feature of the super-pixel region is determined from a sum of horizontal gradient and vertical gradient values of points within the super-pixel region;

the M-dimensional features corresponding to the NIR channel histogram of the super-pixel region are determined according to the following mode:

selecting a histogram with a first preset dimension as M, determining the interval of each dimension, counting the NIR value of the super-pixel region according to the interval, and determining an M-dimensional initial feature; normalizing the M-dimensional initial feature to obtain the M-dimensional feature;

the L-dimensional characteristics corresponding to the histogram of the Gabor filtered channel of the super-pixel region are determined according to the following mode:

and performing filtering operation on the red R, green G and blue B channel images of the super pixel region respectively by using H preset Gabor filters in different directions to obtain corresponding filtering images, and counting the histogram distribution of the super pixel region according to all the filtering images to obtain the L-dimensional features, wherein L is 3M, and H is an integer greater than 1.

In some exemplary embodiments, selecting, according to the water vector data, a region attribute corresponding to a preset number of super-pixel region markers includes:

for each selected super-pixel area, calculating the number of pixels of which the pixels are water bodies in the super-pixel area according to the binarized water body image obtained by converting the water body vector data and the water body range in the super-pixel area;

when the number of the pixels of the water body exceeds a preset proportion of the total number of the pixels of the super pixel area, determining that the area attribute of the super pixel area is the water body; and if the area attribute of the super pixel area is not over, determining that the area attribute of the super pixel area is land.

EXAMPLE five

The embodiment of the present disclosure provides a method for determining a water body extraction model in a remote sensing image, as shown in fig. 6, including:

step 61, obtaining a remote sensing image as a sample;

step 62, acquiring water vector data of a corresponding geographic area from a pre-established water vector data set according to the acquired remote sensing image;

step 63, performing superpixel segmentation on the obtained remote sensing image to obtain a plurality of superpixel regions, and determining the statistical characteristics of each superpixel region;

step 64, selecting area attributes corresponding to a preset number of super-pixel area marks according to the water body vector data, wherein the area attributes comprise: water body, land;

and 65, training a Support Vector Machine (SVM) classifier serving as the water body extraction model according to the regional attributes and the statistical characteristics of the super-pixel regions with the preset number. And the trained SVM classifier is the water body extraction model in the determined remote sensing image.

In some exemplary embodiments, the obtaining, from the obtained remote sensing image (as a sample remote sensing image), water vector data of a corresponding geographic area in a pre-established water vector data set includes:

In some exemplary embodiments, the performing the superpixel segmentation on the obtained remote sensing image (the remote sensing image as the sample) to obtain a plurality of superpixel regions includes:

In some exemplary embodiments, the statistical features include: the method comprises the steps of performing full variation characterization, near infrared spectrum NIR channel histogram characterization of a super pixel region, histogram characterization of a channel after Gabor filtering of the super pixel region and dimensional water body characterization of the super pixel region.

utilizing preset H Gabor filters in different directions to respectively perform filtering operation on the red R, green G and blue B channel images of the super pixel region to obtain corresponding filtering images, and counting the histogram distribution of the super pixel region according to all the filtering images to obtain the L-dimensional features, wherein L is 3M, and H is an integer greater than 1;

taking the super-pixel areas with the area attribute of water in the super-pixel areas of the preset number as positive samples, and taking the super-pixel areas with the area attribute of land in the super-pixel areas of the preset number as negative samples;

training the SVM classifier according to the statistical characteristics of the positive samples and the statistical characteristics of the negative samples; i.e. determining the weights of the SVM classifier.

EXAMPLE six

The embodiment of the present disclosure further provides a water body extraction device in a remote sensing image, including:

the acquisition module is used for acquiring a remote sensing image to be subjected to water body extraction;

the segmentation module is used for carrying out superpixel segmentation on the obtained remote sensing image to obtain a plurality of superpixel regions;

a feature determination module configured to determine a statistical feature for each super-pixel region;

the extraction module is arranged to determine the region attribute of each super-pixel region by utilizing a pre-trained water body extraction model for classification according to the statistical characteristics of each super-pixel region; and marking the area corresponding to the remote sensing image to be subjected to water body extraction as the water body according to the super pixel area determined as the water body, thereby obtaining a final water body extraction result graph.

EXAMPLE seven

The embodiment of the present disclosure provides a device for determining a water body extraction model in a remote sensing image, including:

an acquisition module configured to acquire a remote sensing image as a sample; acquiring water body vector data of a corresponding geographic area in a pre-established water body vector data set according to the acquired remote sensing image;

the characteristic determining module is used for determining the statistical characteristic of each super pixel region;

the training module is configured to select area attributes corresponding to a preset number of super-pixel area markers according to the water body vector data, wherein the area attributes include: water body, land; and training a Support Vector Machine (SVM) classifier serving as the water body extraction model according to the region attributes and the statistical characteristics of the super-pixel regions with the preset number.

Example eight

The embodiment of the present disclosure provides a method for extracting a water body from a remote sensing image, as shown in fig. 7, including,

step 111, acquiring water body vector data from the geographic information of a third-party data source, and establishing a water body vector database, wherein the water body vector database is used for follow-up query; the third-party data source comprises: *** maps, Gade maps, etc. Not limited to geographic information data of a particular company. Acquiring water body vector data from geographic information of a third-party data source according to a related technical method by a person skilled in the art, and establishing a water body vector database; the method for acquiring water vector data in the embodiments of the present disclosure is not limited to a specific manner, and the specific scheme does not belong to the scope defined in the present application.

Step 112, obtaining a remote sensing image as a sample;

step 113, acquiring corresponding geographical area water vector data according to the remote sensing image;

step 114, performing superpixel segmentation on the remote sensing image serving as the sample to obtain a plurality of superpixel regions;

step 115, calculating the statistical characteristics of each super pixel region;

step 116, selecting a preset number of superpixel areas in the obtained remote sensing image as sample superpixel areas, and marking the area attribute of each sample superpixel area according to the obtained geographical area water body vector data; the region attributes include at least: water body and land. And taking the marked super-pixel area as a training sample.

Step 117, training an SVM classifier; namely, training an SVM (Support Vector Machine) classifier according to the statistical characteristics of the training samples; the trained SVM classifier is an extraction model for subsequently extracting the water body, and the SVM comprises relevant weights.

Step 122, obtaining a remote sensing image to be subjected to water body extraction;

step 124, performing superpixel segmentation on the remote sensing image of the water body to be extracted to obtain a plurality of superpixel regions;

step 125, calculating the statistical characteristics of each super-pixel region;

step 126, SVM judgment; namely, the SVM classifier trained in step 117 is used to determine the region attribute of each superpixel region;

step 127, determining a water body extraction result; namely, the super-pixel region of the water body is determined, and the region corresponding to the remote sensing image to be subjected to water body extraction is marked as the water body, so that a final water body extraction result graph is obtained.

In some exemplary embodiments, step 111 is a preparation step performed in advance, and may be performed in advance, and is not limited to being performed with a subsequent classifier training or water body extraction process.

In some exemplary embodiments, the step 112 and 117 is a classifier training process performed by using samples, and the trained classifier may be used for water extraction in multiple remote sensing images, and the training process is not limited to be performed for each water extraction. In some exemplary embodiments, step 112 and 117 are performed in advance.

In some exemplary embodiments, the SVM classifier in step 117 is a linear SVM classifier or a non-linear SVM classifier.

In some exemplary embodiments, step 122-127 is a process of performing water body extraction on the remote sensing image to be subjected to water body extraction. The processing method of the steps 124-125 and the step 114-115 are the same, but the processing objects are different, and the step 114-115 is a step performed on the remote sensing image as the sample.

In some exemplary embodiments, step 112-117 comprises the following steps, as shown in FIG. 2:

step 112 includes: step 1121, obtaining a remote sensing image as a sample;

step 113 comprises: step 1131-133;

step 1131, determining a geographic range corresponding to the remote sensing image;

step 1132, acquiring target water body vector data in a geographic range;

step 1133, converting the target water body vector data into a binary image, and taking the binary image as geographical area water body vector data corresponding to the remote sensing image of the sample;

step 114 comprises: step 1141, performing superpixel segmentation on the remote sensing image serving as a sample by using a Simple Linear Iterative Clustering (SLIC) method to obtain a plurality of superpixel regions;

step 115 comprises: step 1151, calculating the statistical characteristics of each super-pixel region; wherein the statistical features at least include: full variation characteristics, NIR channel histogram characteristics, histogram characteristics after Gabor filtration and water body characteristics;

step 116 includes: step 1161, selecting and marking training samples;

step 117 comprises: step 1171, training an SVM classifier.

In some exemplary embodiments, step 124 includes step 1141 and step 125 includes step 1151, wherein the processed telemetric image is replaced with a telemetric image of the body of water to be extracted.

In some exemplary embodiments, the step 1131 of determining the geographic range corresponding to the remote sensing image includes: and acquiring geographical range information included in the remote sensing image by using a GDAL (geographic Data Abstraction Library), namely a longitude range interval and a latitude range interval corresponding to the remote sensing image.

In some exemplary embodiments, step 1132 obtains target water vector data over a geographic area, including:

In some exemplary embodiments, step 1133 converts the target water vector data into a binary image, including:

In some exemplary embodiments, the segmentation of the remotely sensed image into a plurality of super-pixel regions is performed, in particular, using SLIC (simple linear iterative clustering). The SLIC is a segmentation method which effectively generates compact almost unified superpixels by clustering pixels in a remote sensing image by using the color similarity of the pixels and an image one-plane space. Wherein, step 114 or 1141 includes:

d_cis the color distance, d_sIs the spatial distance, D' is a distance measure, N_cAt a maximum color distance, N_sIs the sampling interval. Converting the image from RGB color space to Lab color space, wherein the (l, a, b) color value and (x, y) coordinate corresponding to each pixel form a 5-dimensional vector V [ l, a, b, x, y ]]The similarity of two pixels can be measured by their vector distance D', the greater the distance, the less similar. j, i respectively represent pixel pointsAnd a seed point.

In some exemplary embodiments, step 115 or 1151 includes:

Wherein the content of the first and second substances,

calculated using the following formula:

4. Determining a final dimension for each super-pixel region, comprising: and calculating the N dimension of each super pixel region, and superposing the dimensions, wherein the dimension is N-1 +16+ 3-6-16-305.

It can be seen that the N-dimensional statistical features of each superpixel region are calculated through steps 1-4 above.

In some exemplary embodiments, step 116 or 1161 includes:

selecting a preset number of super-pixel regions from the plurality of super-pixel regions obtained in step 114 or 1141 as sample super-pixel regions, determining and marking the region attribute of each sample super-pixel region according to the binary image obtained in step 113, where the region attribute label at least includes: water body, land;

In some exemplary embodiments, determining and marking the region attribute of each sample super-pixel region according to the binary image obtained in step 113 includes:

In some exemplary embodiments, step 117 or 1171 comprises:

Example nine

The embodiment of the present disclosure further provides a method for determining a water body extraction model in a remote sensing image, as shown in fig. 7, including:

step 112, obtaining a remote sensing image as a sample;

step 116, selecting and marking training samples;

step 117, training an SVM classifier; and the trained SVM classifier is the determined water body extraction model.

In some exemplary embodiments, further implementation steps of the above steps are similar to corresponding steps in embodiment eight, and are not repeated.

Example ten

The embodiment of the present disclosure further provides a method for extracting a water body from a remote sensing image, as shown in fig. 7, including:

step 126, SVM judgment; determining the region attribute of each super-pixel region by the trained SVM classifier in the eighth or ninth embodiment;

and step 127, determining a water body extraction result.

In some exemplary embodiments, step 126 includes:

In some exemplary embodiments, the pre-trained water body extraction model for classification may be an SVM classifier trained in step 117 in embodiment eight or nine. In some exemplary embodiments, further implementation steps of the above steps are similar to corresponding steps in embodiment eight, and are not repeated.

An embodiment of the present disclosure further provides an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor is configured to execute the computer program to perform any one of the above methods for water body extraction in a remote sensing image.

An embodiment of the present disclosure further provides an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor is configured to execute the computer program to perform any one of the above methods for determining a water body extraction model in a remote sensing image.

The embodiment of the disclosure also provides a storage medium, in which a computer program is stored, where the computer program is configured to execute any one of the above methods for extracting a water body from a remote sensing image when running.

The embodiment of the disclosure also provides a storage medium, in which a computer program is stored, where the computer program is configured to execute the method for determining the water body extraction model in any one of the remote sensing images when running.

It will be understood by those of ordinary skill in the art that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. In a hardware implementation, the division between functional modules/units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be performed by several physical components in cooperation. Some or all of the components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.

Claims

1. A method for extracting water body in remote sensing image is characterized by comprising the following steps,

acquiring a remote sensing image to be subjected to water body extraction;

2. The method of claim 1,

the water body extraction model for classification is trained in advance according to the following method:

obtaining a remote sensing image as a sample;

3. The method of claim 2,

the pre-established water vector data set comprises: and a water body vector database which is established in advance according to the geographic information provided by the third-party geographic information system.

4. The method of claim 2,

the method for acquiring the water body vector data of the corresponding geographic area in the pre-established water body vector data set according to the acquired remote sensing image comprises the following steps:

5. The method according to claim 1 or 2,

the method for performing superpixel segmentation on the obtained remote sensing image to obtain a plurality of superpixel regions comprises the following steps:

6. The method of claim 2,

the training of the SVM classifier as the water body extraction model according to the regional attributes and the statistical characteristics of the super-pixel regions with the preset number comprises the following steps:

7. The method of claim 1,

the determining the region attribute of each super-pixel region by using a pre-trained water body extraction model for classification according to the statistical characteristics of each super-pixel region comprises the following steps:

the following steps are respectively performed for each super pixel region:

8. The method according to claim 1 or 2,

the statistical features include: the method comprises the steps of full variation characteristics, NIR channel histogram characteristics of a near infrared spectrum of a super pixel region and histogram characteristics of a Gabor filtered channel of the super pixel region.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program for water body extraction in a remotely sensed image, and the processor is configured to read and run the computer program for water body extraction in a remotely sensed image to perform the method of any one of claims 1 to 8.

10. A storage medium, in which a computer program is stored, wherein the computer program is arranged to perform the method of any of claims 1 to 8 when executed.