CN115147746A

CN115147746A - Saline-alkali geological identification method based on unmanned aerial vehicle remote sensing image

Info

Publication number: CN115147746A
Application number: CN202211068025.6A
Authority: CN
Inventors: 李伯春
Original assignee: Guangdong Rongqi Intelligent Technology Co Ltd
Current assignee: Guangdong Rongqi Intelligent Technology Co Ltd
Priority date: 2022-09-02
Filing date: 2022-09-02
Publication date: 2022-10-04
Anticipated expiration: 2042-09-02
Also published as: CN115147746B

Abstract

The invention relates to the technical field of image data processing, in particular to a saline-alkali geological identification method based on an unmanned aerial vehicle remote sensing image, which comprises the following steps: acquiring a remote sensing image to be detected of a target ground area under the saline-alkali condition to be detected; carrying out image enhancement extraction processing on a remote sensing image to be detected; carrying out color space conversion processing on the target remote sensing image; carrying out region division optimization on the color remote sensing image; merging and dividing the initial optimal saline-alkali area; cutting each target saline-alkali area; inputting each target saline-alkali image into a saline-alkali degree classification network, and determining the saline-alkali degree corresponding to the target saline-alkali image; and generating target saline-alkali information representing the saline-alkali condition of the target ground area. The method solves the technical problem of low accuracy of saline-alkali identification of the soil by carrying out image processing on the remote sensing image to be detected, improves the accuracy of saline-alkali identification of the soil, and is mainly applied to identification of the saline-alkali degree of the soil.

Description

Saline-alkali geological identification method based on unmanned aerial vehicle remote sensing image

Technical Field

The invention relates to the technical field of image data processing, in particular to a saline-alkali geological identification method based on unmanned aerial vehicle remote sensing images.

Background

Under natural environment influence or artificial movement interference, the salinity in soil and groundwater rises to the earth's surface through the moisture of capillary along with the bottom, and waits to the moisture evaporation after, the salinity often gathers in earth's surface soil, forms saline and alkaline land. Large-scale geological salinization not only affects the land quality in the range, but also often causes the reduction of cultivated land yield and water pollution, even biological diversity. In addition, geological salinization may also cause many other serious land problems, with very serious direct or indirect economic losses to human society. Therefore, the saline-alkali identification of the soil is very important. At present, when carrying out saline-alkali identification to soil, the mode that usually adopts is: firstly, acquiring a soil ground image, inputting the soil ground image into a semantic segmentation network, extracting and segmenting the soil ground image through the semantic segmentation network, finally, respectively inputting a plurality of extracted and segmented images obtained after extraction and segmentation into a classification neural network, and detecting the saline-alkali degree of the plurality of extracted and segmented images through the classification neural network to obtain the saline-alkali degree corresponding to each extracted and segmented image.

However, when the above-described manner is adopted, there are often technical problems as follows:

due to the characteristics of the semantic segmentation network, the relation between pixels is not considered when the soil ground image is extracted and segmented through the semantic segmentation network, the space consistency is lacked, and the extraction and segmentation are not sensitive to details, so that the soil ground image is not accurate enough in extraction and segmentation, and therefore, the extracted and segmented images of regions with various saline-alkali degrees possibly exist in the obtained plurality of extracted and segmented images, the saline-alkali degrees corresponding to the segmented images cannot be extracted through accurate determination of a subsequent classification neural network, and further the saline-alkali identification of soil is low in accuracy.

Disclosure of Invention

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.

In order to solve the technical problem of low accuracy of saline-alkali identification of soil, the invention provides a saline-alkali geological identification method based on an unmanned aerial vehicle remote sensing image.

The invention provides a saline-alkali geological identification method based on an unmanned aerial vehicle remote sensing image, which comprises the following steps:

acquiring a remote sensing image to be detected of a target ground area of a saline-alkali condition to be detected through a target camera installed on a target unmanned aerial vehicle;

carrying out image enhancement extraction processing on the remote sensing image to be detected to obtain a target remote sensing image;

carrying out color space conversion processing on the target remote sensing image to obtain a color remote sensing image;

carrying out region division optimization on the color remote sensing image to obtain a preliminary optimal saline-alkali region set;

merging and dividing the initial optimal saline-alkali regions in the initial optimal saline-alkali region set to obtain a target saline-alkali region set;

cutting each target saline-alkali area in the target saline-alkali area set, and determining a target saline-alkali image corresponding to the target saline-alkali area to obtain a target saline-alkali image set;

inputting each target saline-alkali image in the target saline-alkali image set into a trained saline-alkali degree classification network, and determining the saline-alkali degree corresponding to the target saline-alkali image through the saline-alkali degree classification network;

and generating target saline-alkali information representing the saline-alkali condition of the target ground area according to the saline-alkali degree corresponding to each target saline-alkali image in the target saline-alkali image set.

Further, the region division optimization is performed on the color remote sensing image to obtain a preliminary optimal saline-alkali region set, and the method comprises the following steps:

mapping the color remote sensing image into a target undirected graph;

and determining the initial optimal saline-alkali area set through an optimization algorithm according to the target undirected graph.

Further, the objective function of the optimization algorithm is:

wherein the content of the first and second substances,

is the objective function of the optimization algorithm,Fis the value of the objective function of the optimization algorithm,

is a first objective function of the first set of functions,His the value of the first objective function,

，

so as to make

Counting at the bottom

The number of the pairs is logarithmic,

is the second in the target undirected graphiPixel point and the secondjThe probability of a transition between individual pixel points,

，

，

is the second in the target undirected graphiThe sum of the edge weights between each pixel point and each pixel point in the target undirected graph,Iis the number of pixel points in the target undirected graphiA pixel point and a secondjThe edge weight between each pixel point is

，exp() Is an exponential function with a natural constant as the base,

is the second in the target undirected graphiPixel point and the secondjThe euclidean distance between the individual pixel points,

is the second in the target undirected graphiTone value corresponding to each pixel point and the secondjThe difference in hue value corresponding to each pixel point,

is the second in the target undirected graphiSaturation value corresponding to each pixel point and the secondjThe difference in saturation values corresponding to each pixel point,

is the second in the target undirected graphiBrightness value corresponding to each pixel point and the secondjThe difference in luminance values corresponding to each pixel point,Uis the parameter of the model and is,

is the coefficient of the model and is,

is the second objective functionThe number of the first and second groups is,Objis the value of the second objective function,

is the second in the target undirected graphkThe number of pixels in each sub-region, a sub-region being a region in the target undirected graph,

and

are respectively the second in the target undirected graphkThe length and width of the smallest circumscribed rectangle corresponding to a sub-region,Kis the number of sub-regions into which the target undirected graph is divided.

Further, the preliminary optimal saline-alkali area in the preliminary optimal saline-alkali area set is merged and divided to obtain a target saline-alkali area set, and the method comprises the following steps:

normalizing the target remote sensing image to obtain a target normalized image;

determining a target preliminary region corresponding to each preliminary optimal saline-alkali region in the preliminary optimal saline-alkali region set according to the position of each preliminary optimal saline-alkali region in the color remote sensing image to obtain a target preliminary region set, wherein the target preliminary region in the target preliminary region set is a region in the target normalized image;

dividing a channel value of each channel in a target number of channels of the target normalized image into a preset number of channel levels to obtain a channel level set, wherein each channel of the target normalized image corresponds to the preset number of channel levels, and the number of the channel levels in the channel level set is the power of the preset number of target number;

determining a color channel histogram corresponding to each target preliminary region in the target preliminary region set;

determining a first region merging index between every two target preliminary regions according to the color channel histograms corresponding to the two target preliminary regions in the target preliminary region set and the channel level set;

determining a second region merging index between every two target preliminary regions according to every two target preliminary regions in the target preliminary region set;

determining an overall region merging index between every two target preliminary regions according to a first region merging index and a second region merging index between every two target preliminary regions in the target preliminary region set;

when the overall area merging index between two target preliminary areas in the target preliminary area set is larger than a preset judgment threshold value, dividing the two target preliminary areas into the same type of target preliminary area;

and determining the target preliminary region in the same target preliminary region as a target saline-alkali region in the target saline-alkali region set.

Further, the formula for determining the correspondence between the first region merging indicators of the two target preliminary regions is as follows:

wherein the content of the first and second substances,

is the first in the target preliminary region setpA target preliminary region andqa first region merging indicator between the target preliminary regions,Cis the number of channel levels in the set of channel levels,

is the first in the target preliminary region setpA first one of the channel level sets included in the corresponding color channel histogram for each target preliminary regioncThe histogram distribution values at the level of one channel,

is the first in the target preliminary region setqCorresponding color channel histogram of each target preliminary region first of the channel level sets included thereincThe histogram distribution values at the level of one channel,

is a preset number greater than 0.

Further, the determining a second region merging indicator between every two target preliminary regions according to every two target preliminary regions in the target preliminary region set includes:

translating the two target preliminary regions to enable the distance between the two target preliminary regions to be zero, and obtaining fitting regions corresponding to the two target preliminary regions;

and determining a second region merging index between the two target preliminary regions according to the two target preliminary regions and the corresponding fitting regions of the two target preliminary regions.

Further, the formula for determining the second region merging indicator between the two target preliminary regions is:

wherein the content of the first and second substances,

is the first in the target preliminary region setpA target preliminary region andqsecond region merging indicators between the target preliminary regions,exp() Is an exponential function with a natural constant as the base,

is the first in the target preliminary region setpA target preliminary region andqthe number of pixel points in the fitting and region corresponding to each target preliminary region,

is the first in the target preliminary region setpA target preliminary region andqthe number of edge pixel points included in the fitting and region corresponding to each target preliminary region,

is the first in the target preliminary region setpA target preliminary region andqthe perimeter of the minimum bounding rectangle corresponding to the fitting and region corresponding to each target preliminary region,

is the first in the target preliminary region setpThe number of pixels in the target initial region,

is the first in the target preliminary region setpThe number of edge pixels included in each target preliminary region,

is the first in the target preliminary region setpThe perimeter of the minimum bounding rectangle corresponding to each preliminary region of interest,

is the first in the target preliminary region setqThe number of pixel points within the individual target initialization region,

is the first in the target preliminary region setqThe number of edge pixels included in each target preliminary region,

is the first in the target preliminary region setqThe perimeter of the minimum bounding rectangle corresponding to each target preliminary region.

Further, the formula for determining the whole region merging indicator between the two target preliminary regions is:

wherein the content of the first and second substances,

is the first in the target preliminary region setpA target preliminary region andqthe overall region merging indicators between the target preliminary regions,exp() Is an exponential function with a natural constant as the base,

is the first in the target preliminary region setpA target preliminary region andqa first region merging indicator between the target preliminary regions,

is the first in the target preliminary region setpA target preliminary region andqsecond region merging indicators between the target preliminary regions.

Further, the image enhancement extraction processing is performed on the remote sensing image to be detected to obtain a target remote sensing image, and the image enhancement extraction processing includes:

carrying out image enhancement processing on the remote sensing image to be detected through an image enhancement algorithm to obtain an enhanced image to be detected;

and extracting a target ground area of the enhanced image to be detected to obtain the target remote sensing image.

Further, the training process of the saline-alkali degree classification network comprises the following steps:

constructing a saline-alkali degree classification network;

obtaining a sample saline-alkali image set, wherein a label corresponding to a sample saline-alkali image in the sample saline-alkali image set is the saline-alkali degree;

and training a saline-alkali degree classification network by utilizing the sample saline-alkali image set and labels corresponding to the sample saline-alkali images in the sample saline-alkali image set to obtain the trained saline-alkali degree classification network.

The invention has the following beneficial effects:

according to the saline-alkali geological identification method based on the unmanned aerial vehicle remote sensing image, the technical problem that the accuracy of saline-alkali identification of soil is low is solved by image processing of the remote sensing image to be detected, and the accuracy of saline-alkali identification of the soil is improved. Firstly, a target camera installed on a target unmanned aerial vehicle is used for acquiring a remote sensing image to be detected of a target ground area to be detected for saline-alkali conditions. The remote sensing image to be detected contains the information of the target ground area, so that the subsequent image processing of the remote sensing image to be detected can be facilitated, and the saline-alkali condition corresponding to the remote sensing image to be detected can be determined, so that the saline-alkali condition of the target ground area can be determined, and the accuracy of determining the saline-alkali condition of the target ground area can be improved. And secondly, carrying out image enhancement extraction processing on the remote sensing image to be detected to obtain a target remote sensing image. The image enhancement processing is carried out on the remote sensing image to be detected, the image contrast of the remote sensing image to be detected can be improved, the visualization degree of the remote sensing image to be detected is higher, and the subsequent image processing can be conveniently carried out on the remote sensing image to be detected. Secondly, in actual conditions, through the target camera, not only the target ground area but also an area other than the target ground area is often shot on the obtained remote sensing image to be detected. However, the regions other than the target ground region do not need to be subjected to saline-alkali condition judgment, so that the enhanced image to be detected is extracted to obtain the target remote sensing image only shooting the target ground region, subsequent steps can be omitted in the regions other than the target ground region, the calculation amount can be reduced, the occupation of calculation resources can be reduced, and the efficiency of saline-alkali identification on the target ground region can be improved. And then, carrying out color space conversion processing on the target remote sensing image to obtain a color remote sensing image. Since the color remote sensing image may be an HSV image. HSV images contain three channels, H (Hue), S (Saturation) and V (Value), which tend to be less correlated and more consistent with visual characteristics. Subsequent analysis of the three channels H, S and V may be facilitated. And then, carrying out region division optimization on the color remote sensing image to obtain a preliminary optimal saline-alkali region set. And continuing to merge and divide the initial optimal saline-alkali regions in the initial optimal saline-alkali region set to obtain a target saline-alkali region set. In practical situations, the saline-alkali degree corresponding to each position in the target ground area is often more than one, that is, the target ground area may often include a plurality of areas with different saline-alkali degrees. Therefore, the color remote sensing images are subjected to area division, optimization and combination, a plurality of target saline-alkali areas with different corresponding saline-alkali degrees can be obtained, and the saline-alkali condition of the target ground area can be conveniently and accurately judged subsequently. And then, cutting each target saline-alkali area in the target saline-alkali area set, determining a target saline-alkali image corresponding to the target saline-alkali area, and obtaining a target saline-alkali image set. The different target saline-alkali regions of the corresponding saline-alkali conditions are cut off to obtain the target saline-alkali images corresponding to the saline-alkali degrees one by one, so that the subsequent analysis of the positions of multiple saline-alkali degrees and each saline-alkali degree in the target ground region can be facilitated. Then, inputting each target saline-alkali image in the target saline-alkali image set into a trained saline-alkali degree classification network, and determining the saline-alkali degree corresponding to the target saline-alkali image through the saline-alkali degree classification network. And finally, generating target saline-alkali information representing the saline-alkali condition of the target ground area according to the saline-alkali degree corresponding to each target saline-alkali image in the target saline-alkali image set. Therefore, the method and the device solve the technical problem of low accuracy of saline-alkali identification of the soil by processing the image of the remote sensing image to be detected, and improve the accuracy of the saline-alkali identification of the soil.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a flow diagram of some embodiments of a method for saline-alkali geological identification based on unmanned aerial vehicle remote sensing images according to the invention;

FIG. 2 is a schematic diagram of an enhanced image to be detected and a target remote sensing image according to the present invention;

FIG. 3 is a schematic diagram of a distance between two target preliminary regions being zero according to the present invention;

fig. 4 is a schematic diagram of a target normalized image, a target saline-alkali area and a target saline-alkali image according to the invention.

Wherein the reference numerals in fig. 2 include: the method comprises the steps of an enhanced image to be detected 201, an image area 202 and a target remote sensing image 203.

The reference numerals in fig. 3 include: a first target preliminary region 301 and a second target preliminary region 302.

The reference numerals in fig. 4 include: the target normalized image 401, the first target saline-alkali region 402, the second target saline-alkali region 403, the first target saline-alkali image 404 and the second target saline-alkali image 405.

Detailed Description

To further explain the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the embodiments, structures, features and effects of the technical solutions according to the present invention will be given with reference to the accompanying drawings and preferred embodiments. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

acquiring a remote sensing image to be detected of a target ground area with saline-alkali condition to be detected by a target camera installed on a target unmanned aerial vehicle;

merging and dividing the initial optimal saline-alkali areas in the initial optimal saline-alkali area set to obtain a target saline-alkali area set;

cutting each target saline-alkali area in the target saline-alkali area set, determining a target saline-alkali image corresponding to the target saline-alkali area, and obtaining a target saline-alkali image set;

The following steps are detailed:

referring to fig. 1, a flow of some embodiments of a method for identifying saline-alkali geology based on unmanned aerial vehicle remote sensing images according to the present invention is shown. The saline-alkali geological identification method based on the unmanned aerial vehicle remote sensing image comprises the following steps:

s1, acquiring a remote sensing image to be detected of a target ground area to be detected for saline-alkali conditions through a target camera installed on a target unmanned aerial vehicle.

In some embodiments, the remote sensing image to be detected of the target ground area to be detected for saline-alkali condition can be acquired by a target camera installed on the target unmanned aerial vehicle.

Wherein, above-mentioned target unmanned aerial vehicle is the unmanned aerial vehicle that can normally fly. The target camera may be a remote sensing camera mounted on the target drone. The target ground area may be a ground area to be detected for saline-alkali conditions. The remote sensing image to be detected can be a remote sensing image of a target ground area under the saline-alkali condition to be detected.

As an example, the target unmanned aerial vehicle may be controlled to move to a target position, and a target camera is used to obtain a remote sensing image to be detected of a target ground area under a saline-alkaline condition to be detected. Wherein the target location may be a location directly above the target ground area. The distance between the target position and the target ground area may be a preset height, and may be set according to actual conditions. The shooting angle of the target camera may be a preset angle, and may be set according to actual conditions. The data set according to the actual situation needs to ensure that the remote sensing image to be detected can be acquired through the target camera.

And S2, carrying out image enhancement extraction processing on the remote sensing image to be detected to obtain a target remote sensing image.

In some embodiments, the remote sensing image to be detected may be subjected to image enhancement extraction processing to obtain a target remote sensing image.

The target remote sensing image can be a remote sensing image to be detected after image enhancement extraction processing.

As an example, this step may include the steps of:

firstly, carrying out image enhancement processing on the remote sensing image to be detected through an image enhancement algorithm to obtain an enhanced image to be detected.

The enhanced image to be detected can be a remote sensing image to be detected after image enhancement processing. Image enhancement algorithms may include, but are not limited to: histogram equalization algorithms, laplacian algorithms, and gamma transformation algorithms.

And secondly, extracting a target ground area of the enhanced image to be detected to obtain the target remote sensing image.

The target remote sensing image can be an area image corresponding to a target ground area.

As an example, as shown in fig. 2, a target ground area extraction process may be performed on the enhanced image to be detected 201 to obtain a target remote sensing image 203. The image area 202 may be an area where the target ground area corresponds to the enhanced image 201 to be detected. And cutting the image area 202 from the enhanced image 201 to be detected to obtain a target remote sensing image 203.

In practical situations, through the target camera, not only the target ground area but also an area other than the target ground area is often shot on the obtained remote sensing image to be detected. However, the regions other than the target ground region do not need to be subjected to saline-alkali condition judgment, so that the target ground region extraction processing is performed on the to-be-detected enhanced image to obtain a target remote sensing image only shooting the target ground region, the subsequent steps are not performed on the regions other than the target ground region, the calculation amount can be reduced, the occupation of calculation resources can be reduced, and the efficiency of saline-alkali identification on the target ground region can be improved.

And S3, performing color space conversion processing on the target remote sensing image to obtain a color remote sensing image.

In some embodiments, the target remote sensing image may be subjected to color space conversion processing to obtain a color remote sensing image.

The color remote sensing image can be an HSV image converted from a target remote sensing image.

As an example, the target remote sensing image may be converted into a color remote sensing image by a conventional technique.

And S4, carrying out region division optimization on the color remote sensing image to obtain a preliminary optimal saline-alkali region set.

In some embodiments, the color remote sensing image may be optimized by region division to obtain a preliminary optimal saline-alkali region set.

As an example, this step may comprise the steps of:

and step one, mapping the color remote sensing image into a target undirected graph.

The target undirected graph can be an undirected graph mapped by the color remote sensing image. The nodes of the target undirected graph can be pixel points in the color remote sensing image. The edges of the target undirected graph can be connecting lines among pixel points in the color remote sensing image.

And secondly, determining the initial optimal saline-alkali area set through an optimization algorithm according to the target undirected graph.

Wherein, the saline and alkaline degree that preliminary optimum saline and alkaline region in the above-mentioned preliminary optimum saline and alkaline region set corresponds can be different. The degree of salt/alkali may be, but is not limited to: normal area, light saline-alkali area, moderate saline-alkali area and severe saline-alkali area.

For example, the objective function of the optimization algorithm may be:

wherein, the first and the second end of the pipe are connected with each other,

is the objective function of the optimization algorithm described above.FAre the objective function values of the above-described optimization algorithm.

Is the first objective function.HIs the first objective function value.

。

So as to make

Counting at the bottom

Logarithm. For example,

。

is the second in the above target undirected graphiPixel point and the secondjThe transition probability between individual pixel points.

。

。

Is the second in the above target undirected graphiAnd the sum of the edge weight values between each pixel point and each pixel point in the target undirected graph.IIs the number of pixel points in the target undirected graph. The second in the above objective undirected graphiA pixel point and a secondjThe edge weight between each pixel point is

。exp() Is an exponential function with a natural constant as the base.

Is the second in the above-mentioned target undirected graphiPixel point and the secondjThe Euclidean distance between the pixel points.

Is the second in the above target undirected graphiHue value (H, hue) corresponding to each pixel point and the secondjDifference values of hue values corresponding to the individual pixel points.

Is the second in the above-mentioned target undirected graphiSaturation value (S) corresponding to each pixel point and the second pixel pointjThe difference of the saturation values corresponding to each pixel point.

Is the second in the above target undirected graphiBrightness Value (V, value) corresponding to each pixel point and the second pixel pointjDifference in luminance values corresponding to each pixel point.UAre the model parameters.

Are the model coefficients.

Is the second objective function.ObjIs the second objective function value.

Is the second in the above-mentioned target undirected graphkThe number of pixel points in each sub-region. The sub-region is a region in the above-described target undirected graph.

And

are respectively the second in the above-mentioned target undirected graphkThe length and width of the minimum bounding rectangle corresponding to a sub-region.KIs the number of sub-regions obtained by dividing the target undirected graph. Wherein the model parametersUAnd coefficient of model

May be preset. For example,Uand (5) =2. Model coefficients

May be positive and real.

When the value of the objective functionFAnd when the maximum value is obtained, dividing the obtained area to obtain a preliminary optimal saline-alkali area.

Due to the fact that

。

。

Is the second in the above-mentioned target undirected graphiAnd the sum of the edge weight value between each pixel point and each pixel point in the target undirected graph.IIs the number of pixel points in the target undirected graph. The second in the above objective undirected graphiPixel point and the secondjThe edge weight between each pixel point is

. So that the first objective function valueHEntropy rate values corresponding to the target undirected graph can be characterized. The larger the entropy rate value is, the higher the structural similarity of each sub-region obtained after the color remote sensing image is divided is, namely the greater the similarity between pixel points in the sub-regions is, and the target function value isFThe larger the division tends to be, the better the division results tend to be. Model coefficients

Can be set according to actual conditions, so that the target function

Adding model coefficients

The result can be more in line with the actual situation. Due to the fact that

Is the second in the above target undirected graphkThe number of pixel points in each sub-region. The sub-region is a region in the above-described target undirected graph.

And

are respectively the second in the above-mentioned target undirected graphkThe length and width of the minimum bounding rectangle corresponding to a sub-region. Therefore, it is not only easy to use

Can characterize thekMorphological characteristics of the sub-regions. When the morphological characteristics of the sub-regions are larger, the total number of the divided sub-regionsKThe smaller the number of the sub-regions, the more stable the form of the sub-regions after division, the more compact the structure in the sub-regions, and the objective function valueFThe larger the division result, the better the division result.

And S5, merging and dividing the initial optimal saline-alkali regions in the initial optimal saline-alkali region set to obtain a target saline-alkali region set.

In some embodiments, the preliminary optimal saline-alkali regions in the preliminary optimal saline-alkali region set may be merged and divided to obtain a target saline-alkali region set.

The saline-alkali degree corresponding to each target saline-alkali area in the target saline-alkali area set can be different.

As an example, this step may include the steps of:

firstly, normalizing the target remote sensing image to obtain a target normalized image.

The target normalized image can be a normalized target remote sensing image.

And secondly, determining a target preliminary region corresponding to each preliminary optimal saline-alkali region in the preliminary optimal saline-alkali region set according to the position of each preliminary optimal saline-alkali region in the color remote sensing image, and obtaining a target preliminary region set.

The target preliminary region in the target preliminary region set may be a region in the target normalized image.

For example, the position of the preliminary optimal saline-alkali region in the color remote sensing image can be determined as the target position. And determining the region where the target position in the target normalized image is located as a target preliminary region corresponding to the preliminary optimal saline-alkali region. Wherein, the shape and size of the preliminary optimal saline-alkali region may be the same as the shape and size of the target preliminary region corresponding to the preliminary optimal saline-alkali region.

Because the color remote sensing image is the target remote sensing image after color space conversion, actual scenes corresponding to pixel points at the same position in the color remote sensing image and the target remote sensing image can be the same. The initial optimal saline-alkali region in the initial optimal saline-alkali region set can be a region in the color remote sensing image. The target preliminary regions in the set of target preliminary regions may be regions in the target normalized image. The target normalized image may be a normalized target remote sensing image. Therefore, the actual scenes corresponding to the pixel points at the same position in the color remote sensing image, the target remote sensing image and the target normalized image can be the same. Therefore, the actual scenes corresponding to the image areas at the same position in the color remote sensing image, the target remote sensing image and the target normalized image can be the same. Therefore, the first position may be the same as the second position. The first location may be a location of the preliminary optimal saline-alkali region in the color remote sensing image. The second location may be a location of a preliminary region of interest in the normalized image of interest corresponding to the preliminary optimal saline-alkali region.

And thirdly, dividing the channel value of each channel in the target quantity of channels of the target normalized image into a preset quantity of channel grades to obtain a channel grade set.

The target number may be the number of channels of the target normalized image. For example, the channels of the target normalized image may be an R (Red), G (Green), and B (Blue) channel in RGB mode, respectively. The target number may be 3. Each channel of the target normalized image may correspond to a preset number of channel levels. The preset number may be a preset number. For example, the predetermined number may be 16. The number of channel levels in the set of channel levels may be raised to the power of a predetermined number of target numbers. For example, the number of channel levels in the set of channel levels described above may be 16 to the power of 3.

For example, the preset number may be 2. The channel values of the R channel of the target normalized image may include: 0.1,0.2,0.6 and 0.7. The channel values of the R channel of the target normalized image may be divided into 2 channel levels. Where 0.1 and 0.2 may be one channel level. 0.6 and 0.7 may be another channel rank.

And fourthly, determining a color channel histogram corresponding to each target preliminary region in the target preliminary region set.

The color channel histogram may be a histogram in which the channel level is a horizontal axis value and the number of target pixel points is a vertical axis value. The target pixel points can be pixel points of which the pixel values in the target preliminary region belong to channel levels.

And fifthly, determining a first region merging index between the two target preliminary regions according to the color channel histograms corresponding to each two target preliminary regions in the target preliminary region set and the channel level set.

For example, the above formula for determining the correspondence between the first region merging indicators of the two target preliminary regions may be:

is the first in the set of target preliminary regionspA target preliminary region andqand merging indexes of the first region between the target preliminary regions.CIs the number of channel levels in the set of channel levels described above.

Is the first in the set of target preliminary regionspPreliminary to an objectThe region is in the first channel level set included in the corresponding color channel histogramcHistogram distribution values at each channel level.

Is the first in the set of target preliminary regionsqThe first one of the target preliminary regions in the above channel level set included in the corresponding color channel histogramcHistogram distribution values at each channel level. The histogram distribution value may be a vertical axis value of the color channel histogram.

Is a preset number greater than 0.

May be a very small number.

The effect of (3) is mainly to prevent the denominator from being 0.

Due to the fact that

Is the first in the set of target preliminary regionspThe first one of the target preliminary regions in the above channel level set included in the corresponding color channel histogramcHistogram distribution values at each channel level.

Is the first in the set of target preliminary regionsqThe first one of the target preliminary regions in the above channel level set included in the corresponding color channel histogramcHistogram distribution values at each channel level. Therefore, the temperature of the molten steel is controlled,

can characterize thepA target preliminary region andqthe difference of histogram distribution values of the c-th channel level of the target preliminary region. Since, here, only the determination between the two is requiredDifference does not need to consider positive and negative, so

The difference between the two can be characterized, an

The larger the difference between the two. Therefore, the first and second electrodes are formed on the substrate,

the difference between the p-th target preliminary region and the q-th target preliminary region can be characterized and then squared to obtain

The difference between the p-th target preliminary region and the q-th target preliminary region may be made more conspicuous.

The denominator can be prevented from being 0.

The smaller the size of the tube is,

the larger. Therefore it is firstpA target preliminary region andqfirst region merging indicator between target preliminary regions

The larger, thepA target preliminary region andqthe more similar the individual target preliminary regions, the more likely they can be merged together.

And sixthly, determining a second region merging index between the two target preliminary regions according to every two target preliminary regions in the target preliminary region set.

For example, this step may include the following sub-steps:

and a first substep of translating the two target preliminary regions to make the distance between the two target preliminary regions zero, and obtaining fitting regions corresponding to the two target preliminary regions.

The fitting union region corresponding to the two target preliminary regions may be a region obtained by merging the two target preliminary regions.

For example, when the distance between the two target preliminary regions is greater than zero, the two target preliminary regions are translated to make the distance between the two target preliminary regions zero, and the fitting regions corresponding to the two target preliminary regions are obtained.

For another example, when the distance between the two target preliminary regions is zero, the two target preliminary regions do not need to be translated, and the two target preliminary regions can be directly merged together to obtain a fit region corresponding to the two target preliminary regions. As shown in fig. 3, the distance between the first target preliminary region 301 and the second target preliminary region 302 is zero. The first target preliminary region 301 and the second target preliminary region 302 may be two target preliminary regions.

And a second sub-step of determining a second region merging indicator between the two target preliminary regions according to the two target preliminary regions and the corresponding fitted regions of the two target preliminary regions.

For example, the formula for determining the correspondence between the second region merging indicators of the two target preliminary regions may be:

wherein the content of the first and second substances,

is the first in the set of target preliminary regionspA target preliminary region andqsecond region merging indicators between the target preliminary regions.exp() Is an exponential function with a natural constant as the base.

Is the above-mentioned target preliminary regionThe first in the set of domainspA target preliminary region andqand the number of pixel points in the fitting and region corresponding to each target preliminary region.

Is the first in the set of target preliminary regionspA target preliminary region andqand the number of edge pixel points included in the fitting and region corresponding to the target preliminary region.

Is the first in the set of target preliminary regionspA target preliminary region andqand fitting the corresponding target preliminary regions and determining the circumferences of the minimum bounding rectangles corresponding to the regions.

Is the first in the set of target preliminary regionspThe number of pixel points in each target initial region.

Is the first in the set of target preliminary regionspThe number of edge pixel points included in each target preliminary region.

Is the first in the set of target preliminary regionspThe perimeter of the minimum bounding rectangle corresponding to each target preliminary region.

Is the first in the set of target preliminary regionsqThe number of pixel points in each target initial region.

Is the first in the set of target preliminary regionsqThe number of edge pixel points included in each target preliminary region.

Is the first in the set of target preliminary regionsqThe perimeter of the minimum bounding rectangle corresponding to each target preliminary region.

The closer to 0 the value of (c) is, thepA target preliminary region andqthe more similar the respective target preliminary regions.

Can realize the pair

Normalization is performed to facilitate comparison, an

The larger the size of the tube, the larger the tube,

the smaller, thepA target preliminary region andqsecond region merging indicator between target preliminary regions

And seventhly, determining an overall region merging index between every two target preliminary regions according to the first region merging index and the second region merging index between every two target preliminary regions in the target preliminary region set.

For example, the above formula for determining the correspondence of the whole region combination index between the two target preliminary regions may be:

wherein the content of the first and second substances,

is the first in the set of target preliminary regionspA target preliminary region andqand merging indexes of the whole regions among the target preliminary regions.exp() Is an exponential function with a natural constant as the base.

Is the first in the set of target preliminary regionspA target preliminary region andqand merging indexes of the first areas among the target preliminary areas.

Is the first in the set of target preliminary regionspA target preliminary region andqsecond region merging indicators between the target preliminary regions.

Due to the fact thatpA target preliminary region andqfirst region merging indicator between target preliminary regions

The larger, thepA target preliminary region andqthe more similar the individual target preliminary regions, the more likely they can be merged together. First, thepA target preliminary region andqsecond region merging indicator between target preliminary regions

The larger, thepA target preliminary region andqthe more similar the respective target preliminary regions, the more likely they can be merged together. Therefore it is firstpA target preliminary region andqthe larger the overall region merging index between the target preliminary regions is, thepA target preliminary region andqthe more similar the individual target preliminary regions, the more likely they can be merged together. And is provided with

Realize the pair

Normalization is performed to facilitate comparison, an

The larger the size of the tube is,

the larger. In addition, the overall region merging index integrates the first region merging index and the second region merging index, and the accuracy of determining the overall region merging index is improved.

And eighthly, dividing the two target preliminary regions into the same target preliminary region when the overall region merging index between the two target preliminary regions in the target preliminary region set is larger than a preset judgment threshold value.

The determination threshold may be a minimum overall region merging indicator that considers that the two target preliminary regions are not the same type of target preliminary region. For example, the decision threshold may be 0.75. The same kind of target preliminary region may include target preliminary regions having the same degree of salt and alkali.

And ninthly, determining the target preliminary region in the same target preliminary region as the target saline-alkali region in the target saline-alkali region set.

And S6, cutting each target saline-alkali area in the target saline-alkali area set, determining a target saline-alkali image corresponding to the target saline-alkali area, and obtaining a target saline-alkali image set.

In some embodiments, each target saline-alkali region in the target saline-alkali region set may be cut, and a target saline-alkali image corresponding to the target saline-alkali region is determined, so as to obtain a target saline-alkali image set.

As an example, as shown in fig. 4, a first target saline-alkali area 402 and a second target saline-alkali area 403 in a target normalized image 401 may be subjected to a cutting process, so as to obtain a first target saline-alkali image 404 and a second target saline-alkali image 405. Wherein the first target saline alkali area 402 and the second target saline alkali area 403 can be two target saline alkali areas. The first target saline-alkali image 404 and the second target saline-alkali image 405 may be two target saline-alkali images.

And S7, inputting each target saline-alkali image in the target saline-alkali image set into the trained saline-alkali degree classification network, and determining the saline-alkali degree corresponding to the target saline-alkali image through the saline-alkali degree classification network.

In some embodiments, each target saline-alkali image in the target saline-alkali image set may be input into a trained saline-alkali degree classification network, and the saline-alkali degree corresponding to the target saline-alkali image is determined through the saline-alkali degree classification network.

The saline-alkali degree classification network can be used for judging the saline-alkali degree corresponding to the target saline-alkali image. The saline-alkali degree classification network can be a classification recognition neural network.

As an example, the training process of the salinity classification network may include the following steps:

firstly, a saline-alkali degree classification network is constructed.

This step can be implemented by the prior art, and is not described herein again.

And secondly, acquiring a sample saline-alkali image set.

The sample saline-alkali image in the sample saline-alkali image set can be a ground image with known saline-alkali degree. The label that sample saline and alkaline image in the saline and alkaline image set of above-mentioned sample corresponds can be saline and alkaline degree.

And thirdly, training a saline-alkali degree classification network by using the sample saline-alkali image set and labels corresponding to the sample saline-alkali images in the sample saline-alkali image set to obtain the trained saline-alkali degree classification network.

Wherein, the loss function of the training saline-alkali degree classification network can be a cross entropy loss function.

And S8, generating target saline-alkali information representing the saline-alkali condition of the target ground area according to the saline-alkali degree corresponding to each target saline-alkali image in the target saline-alkali image set.

In some embodiments, the target saline-alkali information representing the saline-alkali condition of the target ground area may be generated according to the saline-alkali degree corresponding to each target saline-alkali image in the target saline-alkali image set.

As an example, the saline-alkali degrees corresponding to the target saline-alkali image in the target saline-alkali image set may be a moderate saline-alkali area and a severe saline-alkali area, respectively. The generated target saline-alkali information for representing the saline-alkali condition of the target ground area can be that the middle part of the target ground area is a severe saline-alkali area, the rest areas are moderate saline-alkali areas, and the saline-alkali degree is relatively severe.

According to the saline-alkali geological identification method based on the unmanned aerial vehicle remote sensing image, the technical problem that the accuracy of saline-alkali identification on soil is low is solved by image processing of the remote sensing image to be detected, and the accuracy of saline-alkali identification on the soil is improved. Firstly, a target camera installed on a target unmanned aerial vehicle is used for acquiring a remote sensing image to be detected of a target ground area to be detected for saline-alkali conditions. Because the remote sensing image to be detected contains the information of the target ground area, the subsequent image processing of the remote sensing image to be detected can be facilitated, and the saline-alkali condition corresponding to the remote sensing image to be detected can be determined, so that the saline-alkali condition of the target ground area can be determined, and the accuracy of determining the saline-alkali condition of the target ground area can be improved. And secondly, carrying out image enhancement extraction processing on the remote sensing image to be detected to obtain a target remote sensing image. The image enhancement processing is carried out on the remote sensing image to be detected, the image contrast of the remote sensing image to be detected can be improved, the visualization degree of the remote sensing image to be detected is higher, and the subsequent image processing can be conveniently carried out on the remote sensing image to be detected. In practical situations, the remote sensing image to be detected obtained by the target camera often shoots not only a target ground area but also areas except the target ground area. However, the regions other than the target ground region do not need to be subjected to saline-alkali condition judgment, so that the enhanced image to be detected is extracted to obtain a target remote sensing image only shooting the target ground region, subsequent steps can be omitted in the regions other than the target ground region, the calculation amount can be reduced, the occupation of calculation resources can be reduced, and the efficiency of saline-alkali identification on the target ground region can be improved. And then, carrying out color space conversion processing on the target remote sensing image to obtain a color remote sensing image. Since the color remote sensing image may be an HSV image. HSV images tend to have less correlation between the three channels of H (Hue), S (Saturation) and V (Value), and tend to be more visually consistent. Subsequent analysis of the three channels H, S and V may be facilitated. And then, carrying out region division optimization on the color remote sensing image to obtain a preliminary optimal saline-alkali region set. And continuing to merge and divide the initial optimal saline-alkali regions in the initial optimal saline-alkali region set to obtain a target saline-alkali region set. In practical applications, the saline-alkali degree corresponding to each position in the target land area is often more than one, that is, the target land area may often include a plurality of areas with different saline-alkali degrees. Therefore, the color remote sensing images are subjected to area division, optimization and combination, a plurality of target saline-alkali areas with different corresponding saline-alkali degrees can be obtained, and the saline-alkali condition of the target ground area can be conveniently and accurately judged subsequently. And then, cutting each target saline-alkali area in the target saline-alkali area set, determining a target saline-alkali image corresponding to the target saline-alkali area, and obtaining a target saline-alkali image set. The different target saline-alkali regions of the corresponding saline-alkali conditions are cut off to obtain the target saline-alkali images in one-to-one correspondence with the saline-alkali degrees, so that the subsequent analysis of the multiple saline-alkali degrees corresponding to the target ground regions and the position of each saline-alkali degree in the target ground regions can be facilitated. Then, each target saline-alkali image in the target saline-alkali image set is input to a trained saline-alkali degree classification network, and the saline-alkali degree corresponding to the target saline-alkali image is determined through the saline-alkali degree classification network. And finally, generating target saline-alkali information representing the saline-alkali condition of the target ground area according to the saline-alkali degree corresponding to each target saline-alkali image in the target saline-alkali image set. Therefore, the method and the device solve the technical problem of low accuracy of saline-alkali identification of the soil by image processing of the remote sensing image to be detected, and improve the accuracy of the saline-alkali identification of the soil.

The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.

Claims

1. A saline-alkali geological identification method based on unmanned aerial vehicle remote sensing images is characterized by comprising the following steps:

2. The method for identifying saline-alkali geology based on unmanned aerial vehicle remote sensing images according to claim 1, wherein the method for carrying out region division optimization on the color remote sensing images to obtain a preliminary optimal saline-alkali region set comprises the following steps:

mapping the color remote sensing image into a target undirected graph;

3. The geological identification method of saltine based on unmanned aerial vehicle remote sensing images as claimed in claim 2, wherein the objective function of the optimization algorithm is:

wherein the content of the first and second substances,

，

is to use

Counting at the bottom

The number of the pairs is logarithmic,

is the second in the target undirected graphiA pixel point and a secondjThe probability of a transition between individual pixel points,

，

，

is the second in the target undirected graphiThe sum of the edge weights between each pixel point and each pixel point in the target undirected graph,Iis the number of pixel points in the target undirected graph, the second in the target undirected graphiPixel point and the secondjThe edge weight between each pixel point is

，exp() Is an exponential function with a natural constant as the base,

is the second in the target undirected graphiA pixel point and a secondjThe Euclidean distance between the pixel points is determined,

is the second in the target undirected graphiEach pixel point corresponds toSaturation value ofjThe difference of the saturation values corresponding to each pixel point,

is the second in the target undirected graphiThe brightness value corresponding to each pixel point isjThe difference of the brightness values corresponding to the individual pixel points,Uis a parameter of the model that is,

is the coefficient of the model and is,

is a function of the second objective function and,Objis the value of the second objective function and,

is the second in the target undirected graphkThe number of pixel points within a sub-region, a sub-region being a region in the target undirected graph,

and

4. The geological recognition method of saline alkali based on the remote sensing image of the unmanned aerial vehicle as claimed in claim 3, wherein the merging and dividing treatment of the initial optimal saline alkali regions in the initial optimal saline alkali region set to obtain a target saline alkali region set comprises:

determining a first region merging index between every two target preliminary regions according to the channel level set and color channel histograms corresponding to the two target preliminary regions in the target preliminary region set;

5. The saltine geological recognition method based on unmanned aerial vehicle remote sensing images as claimed in claim 4, wherein the formula for determining correspondence of the first region merging indicator between the two target preliminary regions is as follows:

wherein the content of the first and second substances,

is the first in the target preliminary region setqA first one of the channel level sets included in the corresponding color channel histogram for each target preliminary regioncThe histogram distribution values at the level of one channel,

is a preset number greater than 0.

6. The saltine geological recognition method based on unmanned aerial vehicle remote sensing images as claimed in claim 4, wherein the determining of the second region merging indicator between each two target preliminary regions according to the two target preliminary regions in the target preliminary region set comprises:

7. The method for identifying the saline-alkali geology based on the unmanned aerial vehicle remote sensing image according to claim 6, wherein the formula for determining the correspondence of the second region merging index between the two target preliminary regions is as follows:

is the first in the target preliminary region setpThe number of pixel points within the individual target initialization region,

8. The saltine geological identification method based on the unmanned aerial vehicle remote sensing image according to claim 4, characterized in that the formula for determining the correspondence of the integral region merging index between the two target preliminary regions is as follows:

9. The method for identifying the saline-alkali geology based on the unmanned aerial vehicle remote sensing image according to claim 1, wherein the step of performing image enhancement extraction processing on the remote sensing image to be detected to obtain a target remote sensing image comprises the following steps:

and extracting and processing the target ground area of the enhanced image to be detected to obtain the target remote sensing image.

10. The method for identifying the saline-alkali geology based on the unmanned aerial vehicle remote sensing image according to claim 1, wherein the training process of the saline-alkali degree classification network comprises the following steps:

constructing a saline-alkali degree classification network;

and training the saline-alkali degree classification network by utilizing the sample saline-alkali image set and the label corresponding to each sample saline-alkali image in the sample saline-alkali image set to obtain the trained saline-alkali degree classification network.