CN108960295B - Multi-temporal fully-polarized SAR image feature extraction method and classification method - Google Patents
Multi-temporal fully-polarized SAR image feature extraction method and classification method Download PDFInfo
- Publication number
- CN108960295B CN108960295B CN201810608408.5A CN201810608408A CN108960295B CN 108960295 B CN108960295 B CN 108960295B CN 201810608408 A CN201810608408 A CN 201810608408A CN 108960295 B CN108960295 B CN 108960295B
- Authority
- CN
- China
- Prior art keywords
- target
- order
- fully
- tensor
- temporal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30188—Vegetation; Agriculture
Abstract
The application provides a method for extracting features of multi-temporal fully-polarized SAR images and a method for classifying vegetation distribution areas. The method for extracting the features of the multi-temporal fully-polarized SAR image comprises the following steps: acquiring a plurality of complete polarization SAR images of different time phases in the same region containing a plurality of types of vegetation; segmenting the multi-temporal fully polarized SAR image to obtain a plurality of targets; creating a multi-order tensor for each target, wherein the first order and the second order in the multi-order tensor represent the row and the column of a coherent matrix or a covariance matrix of the target, and the third order in the multi-order tensor represents different time phases; and reducing the dimension of the multi-order tensor of each target by using a multi-linear principal component decomposition algorithm to obtain the eigenvector of each target. The difference of the vegetation is improved through the feature extraction of the multi-temporal fully-polarized SAR image. The DT classification result shows that the method of the application can obtain higher vegetation identification precision even if the training sample is relatively small.
Description
Technical Field
The application relates to the field of remote sensing application, in particular to a method for extracting features of multi-temporal fully-polarized SAR images and a classification method of the multi-temporal fully-polarized SAR images.
Background
Corn is one of the most important economic crops in the world, is an important component in Chinese agriculture, plays an important role in the agricultural and industrial fields, and therefore, how to effectively monitor the corn is a question worthy of careful consideration.
Accurate images of the earth surface can be obtained through a remote sensing technology, and the analysis result of the remote sensing images can redistribute the land and guarantee the land safety and provide convenience for monitoring the crop growth.
In practice, however, corn is easily confused with other surrounding vegetation due to different farming conditions in various parts of our country. Through the separability comparison of the pixel level and the target level, the backscattering coefficient can not ensure the separability between the corn and other ground objects in the middle and later growth stages of the corn.
Disclosure of Invention
The application provides a method for extracting features of a multi-temporal fully-polarized SAR image; a classification method of vegetation distribution areas of multi-temporal fully-polarized SAR images; the problem of separability between the corn and other ground objects is solved.
In order to solve the above technical problem, an embodiment of the present application provides the following technical solutions:
the application provides a method for extracting features of a multi-temporal fully-polarized SAR image, which comprises the following steps:
acquiring a plurality of complete polarization SAR images of different time phases in the same region containing a plurality of types of vegetation;
segmenting the multi-temporal fully polarized SAR image to obtain a plurality of targets;
creating a multi-order tensor for each target, wherein the first order and the second order in the multi-order tensor represent the row and the column of a coherent matrix or a covariance matrix of the target, and the third order in the multi-order tensor represents different time phases;
and reducing the dimension of the multi-order tensor of each target by utilizing a multi-linear principal component decomposition algorithm to obtain the eigenvector of each target.
Optionally, the coherence matrix or covariance matrix includes: intensity information and/or phase information in fully polarised SAR images of several different phases of the same region.
Optionally, before the segmenting the multi-temporal fully polarized SAR image, the method further includes:
and registering the multi-temporal fully polarized SAR images, and performing mean filtering of a preset specification.
Optionally, the step of segmenting the multi-temporal fully-polarized SAR image to obtain a plurality of targets includes:
calculating the distance measure of the super-pixel segmentation by using the average coherent matrix or the average covariance matrix of the multi-temporal fully-polarized SAR image;
aggregating pixels of the multi-temporal fully-polarized SAR image into a plurality of targets by using a distance measure of superpixel segmentation;
and the center of the target is arranged on the pixel with the lowest gradient in the neighborhood of the preset specification.
Optionally, the step of using a multi-linear principal component decomposition algorithm to perform dimensionality reduction on the multi-order tensor of each target to obtain the eigenvector of each target includes:
obtaining a projection matrix from the multi-order tensors of each target
Wherein L is n <I n ， L n Is the dimensionality after dimensionality reduction, I n Is the original dimension;
and reducing the dimension of the multi-order tensor of each target based on the iterative computation of the projection matrix to obtain the eigenvector of each target.
Further, the projection matrix
Wherein L is n <I n ，L n Is the dimensionality after dimensionality reduction, I n For the original dimension, the specific formula is as follows:
when the temperature is higher than the set temperature
When the temperature of the water is higher than the set temperature,
wherein the content of the first and second substances,
i represents the ith order, and n represents the order;
m denotes an mth target tensor, M denotes the number of target tensors;
the value range of Q is [0, 1], and the value range is a user-defined parameter and is used for controlling the ratio of the information stored after the original tensor is projected.
Further, the step of performing dimensionality reduction on the multi-order tensor of each target based on the iterative computation of the projection matrix to obtain the eigenvector of each target includes:
acquiring a plurality of sample targets expressed by tensors;
performing multi-linear projection calculation on the sample target to obtain a tensor Y;
iteratively calculating based on the projection matrix and the tensor Y;
and when the total divergence of the tensor Y converges to meet a preset condition or the iteration times reach a preset value, acquiring the eigenvector of each target.
Optionally, the multi-order tensor further includes a fourth order and a fifth order, which represent the length and width in the neighborhood of the target window.
The application provides a method for classifying vegetation distribution areas of multi-temporal fully-polarized SAR images, which comprises the following steps:
extracting feature vectors of each target of a plurality of different time phases of the fully-polarized SAR images of the same region containing a plurality of vegetation by using the method;
and determining the vegetation class to which the target belongs by utilizing the classifier based on the feature vector of at least one time phase of each target.
Optionally, the average coherence matrix or the average covariance matrix of the target is represented as:
wherein, the first and the second end of the pipe are connected with each other,
T i a coherence matrix representing the ith pixel;
k denotes that the object is made up of k pixels.
Based on the disclosure of the above embodiments, it can be known that the embodiments of the present application have the following beneficial effects:
the application provides a method for extracting features of a multi-temporal fully polarized SAR image and a method for classifying vegetation distribution areas of the multi-temporal fully polarized SAR image. The method for extracting the features of the multi-temporal fully-polarized SAR image comprises the following steps: acquiring a plurality of complete polarization SAR images of different time phases in the same region containing a plurality of types of vegetation; segmenting the multi-temporal fully polarized SAR image to obtain a plurality of targets; creating a multi-order tensor for each target, wherein the first order and the second order in the multi-order tensor represent the row and the column of a coherent matrix or a covariance matrix of the target, and the third order in the multi-order tensor represents different time phases; and reducing the dimension of the multi-order tensor of each target by utilizing a multi-linear principal component decomposition algorithm to obtain the eigenvector of each target. And determining the vegetation type of each target by using the classifier based on the characteristic vector of at least one time phase of each target. The difference of vegetation is improved through the characteristic extraction of the multi-temporal fully-polarized SAR image. The DT classification result shows that the method can obtain higher vegetation identification precision even if the training sample is relatively small. In particular, 3-phase data combination is the optimal choice.
Drawings
Fig. 1 is a flowchart of a method for extracting features from multi-temporal fully-polarized SAR images according to an embodiment of the present application;
fig. 2 is a flowchart of a method for classifying vegetation distribution areas of a multi-temporal fully-polarized SAR image according to an embodiment of the present application.
Detailed Description
Specific embodiments of the present application will be described in detail below with reference to the accompanying drawings, but the present application is not limited thereto.
It will be understood that various modifications may be made to the embodiments disclosed herein. Accordingly, the foregoing description should not be considered as limiting, but merely as exemplifications of embodiments. Those skilled in the art will recognize other modifications that are within the scope and spirit of the present application.
The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the application and, together with a general description of the application given above, and the detailed description of the embodiments given below, serve to explain the principles of the application.
These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment, given as a non-limiting example, with reference to the attached drawings.
It is also understood that, although the present application has been described with reference to some specific examples, a person of skill in the art shall certainly be able to achieve many other equivalent forms of application, having the characteristics as set forth herein and hence all coming within the field of protection defined thereby.
The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.
Specific embodiments of the present application are described hereinafter with reference to the drawings; however, it is to be understood that the disclosed embodiments are merely examples of the application, which can be embodied in various forms. Well-known and/or repeated functions and constructions are not described in detail to avoid obscuring the application of unnecessary or unnecessary detail. Therefore, specific structural and functional details disclosed herein are not intended to be limiting, but merely as a basis for the description and as a representative basis for teaching one skilled in the art to variously employ the present application in virtually any appropriately detailed structure.
The specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," which may each refer to one or more of the same or different embodiments in accordance with the application.
China is facing serious challenges to meet the food needs of nearly 14 million people, and concerns about food production and safety are also rising. Corn plays an important role in both agricultural and industrial fields as one of the most important grains worldwide, and therefore, how to effectively monitor corn is a matter worth careful consideration.
However, due to different farming conditions in different parts of our country, corn is easily confused with other peripheral vegetation. Through the separability comparison of the pixel level and the target level, the separability between the corn and other ground objects cannot be guaranteed by the existing monitoring method in the middle and later growth stages of the corn.
The application provides a method for extracting features of a multi-temporal fully-polarized SAR image; the application also provides a multi-temporal classification method for vegetation distribution areas of the full-polarization SAR image. Details are described in the following examples one by one.
The first embodiment provided by the present application, that is, an embodiment of a method for extracting features for a multi-temporal fully-polarized SAR image.
The present embodiment is described in detail below with reference to fig. 1, where fig. 1 is a flowchart of a method for extracting features from multi-temporal fully-polarized SAR images according to an embodiment of the present application.
Step S101, acquiring a plurality of complete polarization SAR images of different time phases in the same region containing a plurality of types of vegetation.
Synthetic Aperture Radar (SAR for short) is a high-resolution imaging Radar, can obtain high-resolution Radar images similar to optical photography under meteorological conditions with extremely low visibility, and has monitoring advantages compared with optical remote sensing because SAR can overcome the influence of weather and time.
The SAR image is an image obtained by SAR.
Polarization is one of the essential attributes of electromagnetic waves, is important information of another dimension except frequency, amplitude and phase, and completely embodies the vectorial characteristics of the electromagnetic waves. As the polarized SAR (polsar) technology becomes more mature, the amount of information that SAR can provide is also richer. SAR often uses four polarimetric formats-HH, VV, HV, VH. The first two are codirectional polarized and the second two are heterodromous (cross) polarized. The fully polarimetric sar (fp sar) technique has the highest difficulty, can provide intensity and coherent phase information of 4 channels, and requires to transmit H and V simultaneously, i.e. four polarimetric patterns of HH, HV, VV, VH.
The fully polarized SAR image is image information obtained by the SAR through the strength and coherent phase information of 4 channels.
Multi-temporal generally refers to a feature that reflects a set of remote-sensed images in a time series. Broadly speaking, all a set of images, maps or geographic data of the same region, which are acquired at different times, can be regarded as "multi-temporal" data.
The plurality of completely polarized SAR images in different time phases in the same region are completely polarized SAR images in different time phases in the same region. For example, in the Wuqing district (center coordinates of about 117 ° 2 '2 "E, 39 ° 28' 29" N) of Tianjin, which is located in North China plain, is one of the most prominent corn producing areas in China. Three images of the corn in the region, such as DOY207, DOY231 and DOY255 images, are obtained through the full polarization SAR.
The term "day of year" (DOY for short) is a method for continuously calculating dates used only in one year, and is the number of days counted from 1 month to 1 day of the year. For example: DOY207 is the 207 th day from 1 month 1 of the year.
And the registration refers to the matching of geographic coordinates of different image graphs obtained by different imaging means in the same area. The method comprises three processes of geometric correction, projective transformation and unified scale.
The registration of the multi-temporal fully-polarized SAR images is to obtain fully-polarized SAR images in different time periods in the same region through the fully-polarized SAR, select one of the images in one time period as a standard, and perform three aspects of geometric correction, projection transformation and unified scale on other images. For example, if images of three views DOY207, DOY231, DOY255 of corn are obtained in wuqing zone of Tianjin (center coordinates of about 117 ° 2 '2 "E, 39 ° 28' 29" N), then the images of DOY231 and DOY255 may be registered using the image of DOY207 as a standard.
The mean filtering is a typical linear filtering algorithm, which means that a template is given to a target pixel on an image, the template includes neighboring pixels around the target pixel (for example, a filtering template with a size of 3 × 3 is a filtering template that includes 9 pixels and is formed by adding 8 pixels around the target pixel as a center to the target pixel), and the original pixel value is replaced by an average value of all pixels in the template. The purpose of the mean filtering adopted in this embodiment is to reduce the speckle noise, and at the same time, avoid losing the heterogeneity of the polarization features in different acquisition times, which plays an important role in accurately identifying the vegetation.
The performing of the average filtering of the preset specification is to select a window of the preset specification as a sliding window, and perform the average filtering on the image of the window, where the preset specification is 5 × 5, for example.
And S102, segmenting the multi-temporal fully polarized SAR image to obtain a plurality of targets. The method comprises the following specific steps:
and S102-1, calculating the distance measure of the super-pixel segmentation by using the average coherence matrix or the average covariance matrix of the multi-temporal fully-polarized SAR image.
The super-pixel is a small area which is composed of a series of pixel points with adjacent positions and similar characteristics such as color, brightness, texture and the like. Most of these small regions retain effective information for further image segmentation, and generally do not destroy the boundary information of objects in the image.
Superpixel segmentation, in the field of computer vision, refers to the process of subdividing a digital image into a plurality of image sub-regions (sets of pixels), also called superpixels.
Distance measure, i.e. measure the distance measure between different objects. In the embodiment, the distance between pixels is represented by using the concept of distance measure in geometric measurement, and the smaller the distance measure between two pixels is, the higher the probability that the two pixels belong to the same class is.
The distance measure for calculating the super-pixel segmentation is just the difference value of color values between pixels in a common image or an optical image, and the areas of different pixels in the image can be effectively divided through the difference value. In this embodiment, the distance measure of the super-pixel segmentation is used to replace the euclidean distance of the original color space, and the SLIC algorithm is applied to the FP SAR image.
In order to simplify the operation, in the embodiment, the distance measure of the super-pixel segmentation is calculated by using the average coherence matrix or the average covariance matrix of the multi-temporal N-view image fully-polarized SAR image.
A distance measure of the superpixel segmentation, comprising: wishart distance measure, cartesian coordinate distance measure;
calculating a Wishart distance measure D by using an average coherence matrix or a covariance matrix of a multi-temporal N-scene image full-polarization SAR image:
wherein the content of the first and second substances,
d w (i, j) represents the Wishart distance between the ith pixel and the jth target center; further, the center of the target is arranged on the pixel with the lowest gradient in the neighborhood of the preset specification, so that the center of the target is prevented from being located on the boundary;
T i a coherence matrix or covariance matrix representing the ith pixel; further, the coherenceThe matrix or covariance matrix comprises: intensity information and/or phase information in a plurality of different time phases of the same region of the fully-polarized SAR image;
Σ j an average coherence matrix or an average covariance matrix representing the jth target;
d s representing the spatial distance between the ith pixel and the jth target center, i.e. a Cartesian coordinate distance measure, its coordinates (r) i ，c i ) And (r) j ，c j ) Respectively representing the row and column positions of the pixels on the SAR image;
the weighting parameter m and the area parameter S are customized by a user, for example, N is 3, the weighting parameter m is 1, and the area parameter S is 15.
And step S102-2, aggregating the pixels of the multi-temporal fully-polarized SAR image into a plurality of targets by using the distance measure of the super-pixel segmentation.
Step S103, creating a multi-order tensor for each target, wherein the first order and the second order in the multi-order tensor represent the rows and the columns of a coherent matrix or a covariance matrix of the target, and the third order in the multi-order tensor represents different time phases.
Zhang Liang
Can be regarded as an n-dimensional array, n representing the order. The tensor expression method avoids a vectorization process which is usually carried out in the process of processing the multi-temporal N scene image full polarization SAR image, and keeps the data structure of a coherent matrix.
For polarized SAR data, the coherence matrix and covariance matrix are both 3 × 3Hermitian semi-positive matrices, which belong to the riemann manifold and can therefore be considered to be in the form of a second order tensor. Further, the coherence matrix or covariance matrix comprises: intensity information and phase information in a plurality of different time phases of the same region of the fully-polarized SAR image. For example, for a fully polarized SAR image of several different phases of corn, each pixel can be represented as a third order tensor. The coherence matrix and the covariance matrix are both 3 × 3Hermitian semi-positive definite matrices, so the first order and the second order are both 3, and in this embodiment, the third order is 3 because 3 full-polarization SAR images with different time phases are selected.
In fully polarized SAR images of more different phases, the order of the tensor can be simply increased. For example, in order to protect the neighborhood of data and the texture information without superpixel segmentation, the multi-order tensor may further include a fourth order and a fifth order, which represent the length and width in the neighborhood of the target window.
And S104, reducing the dimension of the multi-order tensor of each target by using a multi-linear principal component decomposition algorithm to obtain the eigenvector of each target.
A Multilinear Principal Component Analysis (MPCA) is a tensor-tensor projection method, and can reduce the feature dimension by extracting the original information of the core tensor.
Step S104-1, obtaining a projection matrix according to the multi-order tensor of each target
Wherein L is n <I n ，L n Is the dimensionality after dimensionality reduction, I n Is the original dimension. The specific formula is as follows:
when the temperature is higher than the set temperature
When the temperature of the water is higher than the set temperature,
wherein the content of the first and second substances,
1)
2)
3) i represents the ith order, and n represents the order;
4) m denotes an mth target tensor, M denotes the number of target tensors;
5) the value range of Q is [0, 1], and the value range is a user-defined parameter and is used for controlling the ratio of the information stored after the original tensor is projected.
6) Symbol(s)
Representing the Frobenius norm. For example, for the tensor
The Frobenius norm of (A) can be expressed in the form in which
Represents the inner product of two tensors:
7) matrix U (i) From a matrix phi (i) Maximum L i The feature vector corresponding to each feature value comprises:
8) dimension L i By a matrix
Is determined. Will matrix
Characteristic value of
Arranged in descending order and the calculator accumulates the probabilities
Then L is i Corresponding to the satisfaction of p t <The maximum t of Q.
And S104-2, reducing the dimension of the multi-order tensor of each target based on the iterative computation of the projection matrix to obtain the eigenvector of each target.
The specific algorithm is as follows:
in step S104-2-1, a plurality of sample objects expressed by tensors are acquired.
In specific applications, when features are extracted for multi-phase fully-polarized SAR images, some samples are often selected from targets generated by the fully-polarized SAR images of different phases, and the features are extracted from the samples. The sample target tensor set may be represented as X ═ { X ═ X 1 ,X 2 ,…,X M Where M denotes the number of sample objects. For example, the sample object tensor of maize is concentrated, X i ∈R 3 ×3×N (ii) a Representing the ith sample object.
And step S104-2-2, performing multi-linear projection calculation on the sample target to obtain a tensor Y.
Y＝X× 1 U (1) × 2 U (2) …× n U (n) ； (3)
Wherein the tensor
For the target tensor sample set X ═ X 1 ，X 2 ，…，X M Multiple linear projection results of.
And step S104-2-3, performing iterative computation based on the projection matrix and the tensor Y.
Matrix U (i) By using
Maximum L i The feature vectors are used as initial values, and the sequential iterative calculations of expressions (2), (3), and (1) are performed by an alternating least square algorithm.
And S104-2-4, when the total divergence of the tensor Y is converged to meet a preset condition or the iteration times reach a preset value, acquiring the eigenvector of each target.
For example, the preset condition is
The preset value is 4, and the k value is the current iteration number.
Based on the specific application of the first embodiment provided by the application, the application also provides a second embodiment, namely a method for classifying vegetation distribution areas of multi-temporal fully-polarized SAR images. Since the second embodiment is based on the first embodiment, the description is relatively simple, and the relevant portions only need to refer to the corresponding description of the first embodiment. The second embodiment described below is merely illustrative.
Fig. 2 shows an embodiment of a method for classifying vegetation distribution areas of a multi-temporal fully-polarized SAR image provided by the present application. Fig. 2 is a flowchart of a method for classifying vegetation distribution areas of a multi-temporal fully-polarized SAR image according to an embodiment of the present application.
Referring to fig. 2, the present application provides a method for classifying vegetation distribution areas of a multi-temporal fully-polarized SAR image, including:
the method according to the first embodiment extracts feature vectors of respective targets of a number of different phases of a fully-polarized SAR image of the same region containing a number of vegetation.
Steps 101, 102, 103 and 104 and their sub-steps of the method described in the first embodiment correspond to steps 201, 202, 203 and 204 and their sub-steps, respectively, of this embodiment.
According to the first embodiment, since the order of the output tensor is determined by the parameter Q and represents the ratio of the eigenvalues of the projection matrix to the original information, different sample object sets expressed by the tensor do not necessarily obtain the output result with the same dimension even if they have the same Q value. For example, in the study area, the order of the sample target set expressed in tensor is rapidly decreased from 27(3 × 3 × 3) orders to 12(2 × 2 × 3) orders and 8 orders (2 × 2 × 2). When Q is 0.82, the output tensors are all 4(2 × 2 × 1) orders.
In this example, a method of classifying the distribution region of the 7 types of features was mainly studied. The 7 types of ground objects are: 1, corn; 2, soybean; 3, rice; 4, lotus; 5 grassland; 6, water body; 7, building. The purpose of the study was: classification of vegetation distribution regions of multi-temporal fully-polarized SAR images based on tensor representations.
For accurate identification of corn growing areas, the texture information of the plot needs to be as uniform as possible. Thus, the average coherence matrix or average covariance matrix of the object is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
T i a coherence matrix representing the ith pixel;
k denotes that the object is made up of k pixels.
Researches show that the separability of corn-rice, corn-lotus and corn-grassland is very obvious. Both maize and soybean have well-defined clusters, but both have some overlap in all subspaces, indicating that the polarization characteristics of maize and soybean have some similarity in the mid-late stages of growth. The JM distance between corn and soybean can only reach 1.7 when all three images are used, and the separability is affected to a certain extent because the information loss caused by dimensionality reduction is particularly suffered. However, if the samples of the water body and the building are also added into the tensor set, the projection model obtained by the MPCA algorithm is also changed correspondingly, and the separability between the corn and the soybean is improved to a certain degree.
Furthermore, the difference between the corn and the soybean on a single image is not obvious, and partial subtle differences exist in the change of different time phases, so that useful information can be provided for the middle and later stage charting of the corn.
The method of the first embodiment is continued with the steps of:
step S205, based on the feature vector of at least one time phase of each target, determines the vegetation type to which the target belongs by using a classifier.
After feature dimension reduction, classification can be completed by using a multi-linear classifier, and the DT classifier with the higher speed is selected in the embodiment.
The training samples contain all 7 classes of ground features, and the parameter Q of the MPCA algorithm is set to 0.95, so that the 4(2 x 1) dimensional reduced feature is obtained. The training data set of this embodiment includes 1758 samples of 7 classes of terrain.
A number of samples were randomly selected from the training data set and their average accuracy, including overall accuracy, kappa coefficient, user accuracy and producer accuracy for corn, was calculated. The producer precision and the user precision of the corn are improved when the proportion of the training samples is increased from 10% to 20%, and the producer precision and the user precision of the corn fluctuate around the mean values 0.9388 and 0.9533 respectively by adding the training samples.
At a training sample rate of 10%, the overall accuracy, kappa factor and corn producer accuracy of the DOY231 data were significantly higher than those of DOY207 and DOY255, demonstrating that the heading date data were better able to characterize the corn. However, the data combination of DOY231 and DOY255 improves corn identification results. The producer accuracy of corn is significantly improved compared to the data combination of a single SAR image and DOY207 and 231, while the user accuracy of corn is over 0.95. The 3-phase data set has the highest overall precision, kappa coefficient, and corn producer precision.
On the other hand, when the training sample ratio is 50%, the classification accuracy of DOY255 is equivalent to that of DOY231, and both are significantly better than those of DOY207 data. The data combination of DOY231 and 255 achieves the highest overall accuracy and kappa coefficient, but the 3-phase data combination also achieves a comparable level with a slightly higher accuracy for the corn producer. The overall accuracy and kappa coefficient of the DOY207 and DOY255 data were significantly improved compared to the 10% training sample ratio, but in contrast to DOY255, the corn classification results for DOY207 were not significantly improved. Both the corn producer and user precision of the DOY255 data, and the producer precision of the DOY231 data, improved with the increase in training samples. This indicates that sufficient training samples are beneficial for corn mapping. However, when the input features themselves are slightly less effective, such as the DOY207 data, the addition of training samples does not provide a significant improvement. The same conclusions can be drawn from the multi-temporal data as well, and for the data combinations of DOYs 207 and 231, the increase in training samples only leads to an improvement in the accuracy of the corn user, while for the 3-temporal combination and the data combinations of DOYs 231 and 255, both the corn recognition rate and the overall performance of the classification are improved.
In order to study the effectiveness of the polarized SAR to the mapping of the corn in the middle and later stages of vegetation growth, the embodiment provides a method for classifying vegetation distribution areas of a multi-temporal fully polarized SAR image based on tensor expression, and feature extraction is realized through an MPCA algorithm based on tensor. This embodiment gives a general tensor representation of the coherence matrix of a multi-temporal fully polarimetric SAR image. The present document compares the corn and soybean polarization responses and further emphasizes that the multi-temporal phase change of the polarization information helps to improve the corn-soybean diversity. The DT classification results show that the processing procedure proposed herein can achieve higher corn identification accuracy even in the case of relatively small training samples. Moreover, heading date data is more favorable for corn identification than the jointing and maturation stages, although 3-phase data combinations remain the best choice.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the electronic device to which the data processing method described above is applied may refer to the corresponding description in the foregoing product embodiments, and details are not repeated herein.
The above embodiments are only exemplary embodiments of the present application, and are not intended to limit the present application, and the protection scope of the present application is defined by the claims. Various modifications and equivalents may be made by those skilled in the art within the spirit and scope of the present application and such modifications and equivalents should also be considered to be within the scope of the present application.
Claims (9)
1. A method for extracting features for a multi-temporal fully polarimetric SAR image, the method comprising the steps of:
acquiring a plurality of complete polarization SAR images of different time phases in the same region containing a plurality of types of vegetation;
segmenting the multi-temporal fully polarized SAR image to obtain a plurality of targets, wherein the distance measure of super-pixel segmentation is calculated by utilizing an average coherence matrix or an average covariance matrix of the multi-temporal fully polarized SAR image;
aggregating pixels of the multi-temporal fully-polarized SAR image into a plurality of targets by using a distance measure of superpixel segmentation;
the center of the target is arranged on a pixel with the lowest gradient in the neighborhood of a preset specification;
creating a multi-order tensor for each target, wherein the first order and the second order in the multi-order tensor represent the row and the column of a coherent matrix or a covariance matrix of the target, and the third order in the multi-order tensor represents different time phases;
and reducing the dimension of the multi-order tensor of each target by utilizing a multi-linear principal component decomposition algorithm to obtain the eigenvector of each target.
2. The method of claim 1, wherein the coherence matrix or covariance matrix comprises: intensity information and/or phase information in fully polarized SAR images of several different phases of the same region.
3. The method of claim 1, further comprising, prior to said segmenting the multi-phase fully polarized SAR image:
and registering the multi-temporal fully polarized SAR images, and performing mean filtering of a preset specification.
4. The method of claim 1, wherein the step of using a multi-linear principal component decomposition algorithm to reduce the dimension of the multi-order tensor of each object to obtain the eigenvector of each object comprises:
obtaining a projection matrix from the multi-order tensors of each object
Wherein L is n <I n ，L n Is the dimensionality after dimensionality reduction, I n Is the original dimension;
and reducing the dimension of the multi-order tensor of each target based on the iterative computation of the projection matrix to obtain the eigenvector of each target.
5. The method of claim 4, wherein the projection matrix
Wherein L is n <I n ，L n Is the dimensionality after the dimensionality reduction, I n For the original dimension, the specific formula is as follows:
when in use
When the utility model is used, the water is discharged,
wherein the content of the first and second substances,
i represents the ith order, and n represents the order;
m denotes an mth target tensor, M denotes the number of target tensors;
the value range of Q is [0, 1], and the value range is a user-defined parameter and is used for controlling the ratio of the information stored after the original tensor is projected.
6. The method of claim 5, wherein the step of performing dimensionality reduction on the multiple-order tensor for each target based on the iterative computation of the projection matrix to obtain the eigenvector for each target comprises:
acquiring a plurality of sample targets expressed by tensors;
performing multi-linear projection calculation on the sample target to obtain a tensor Y;
iteratively calculating based on the projection matrix and the tensor Y;
and when the total divergence of the tensor Y converges to meet a preset condition or the iteration times reach a preset value, acquiring the eigenvector of each target.
7. The method of claim 1, further comprising a fourth order and a fifth order of the multi-order tensor, representing a length and a width in a neighborhood of the target window.
8. A method for classifying vegetation distribution areas of multi-temporal fully-polarized SAR images is characterized by comprising the following steps:
extracting feature vectors of respective targets of a number of different time phases of a fully polarised SAR image of the same region containing a number of vegetation using the method of any of claims 1-7;
and determining the vegetation class to which the target belongs by utilizing the classifier based on the feature vector of at least one time phase of each target.
9. The classification method according to claim 8, characterized in that the mean coherence matrix or mean covariance matrix of the object is expressed as:
wherein the content of the first and second substances,
T i a coherence matrix representing the ith pixel;
k denotes that the object is made up of k pixels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810608408.5A CN108960295B (en) | 2018-06-13 | 2018-06-13 | Multi-temporal fully-polarized SAR image feature extraction method and classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810608408.5A CN108960295B (en) | 2018-06-13 | 2018-06-13 | Multi-temporal fully-polarized SAR image feature extraction method and classification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108960295A CN108960295A (en) | 2018-12-07 |
| CN108960295B true CN108960295B (en) | 2022-08-26 |
Family
ID=64488441
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810608408.5A Active CN108960295B (en) | 2018-06-13 | 2018-06-13 | Multi-temporal fully-polarized SAR image feature extraction method and classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108960295B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110399832B (en) * | 2019-07-25 | 2021-08-13 | 内蒙古工业大学 | TomosAR vegetation pest and disease damage monitoring method and device based on coherence |
| CN112749705A (en) * | 2019-10-31 | 2021-05-04 | 深圳云天励飞技术有限公司 | Training model updating method and related equipment |
| CN111523567B (en) * | 2020-03-31 | 2023-09-01 | 彩虹无人机科技有限公司 | Polarization feature extraction and classification method based on tensor local preservation projection |
| CN111967537B (en) * | 2020-04-13 | 2022-04-08 | 江西理工大学 | SAR target classification method based on two-way capsule network |
| CN112257515B (en) * | 2020-09-29 | 2022-09-23 | 河南大学 | SAR image terrain classification method and device based on complex terrain |
| CN113642399A (en) * | 2021-07-12 | 2021-11-12 | 广东省国土资源测绘院 | Method for identifying abandoned land in paddy field based on SAR data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700110A (en) * | 2015-04-03 | 2015-06-10 | 电子科技大学 | Plant covering information extracting method based on perfect polarization SAR images |
| CN104751183A (en) * | 2015-04-08 | 2015-07-01 | 西安电子科技大学 | Polarimetric SAR image classification method based on tensor MPCA |
| CN105930846A (en) * | 2016-04-05 | 2016-09-07 | 西安电子科技大学 | Neighborhood information and SVGDL (support vector guide dictionary learning)-based polarimetric SAR image classification method |
| CN107239757A (en) * | 2017-05-23 | 2017-10-10 | 西安电子科技大学 | A kind of polarization SAR silhouette target detection method based on depth ladder net |
| CN107292920A (en) * | 2017-06-15 | 2017-10-24 | 武汉大学 | A kind of multidate full-polarization SAR remote sensing imagery change detection method of joint classification |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9239384B1 (en) * | 2014-10-21 | 2016-01-19 | Sandia Corporation | Terrain detection and classification using single polarization SAR |
-
2018
- 2018-06-13 CN CN201810608408.5A patent/CN108960295B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700110A (en) * | 2015-04-03 | 2015-06-10 | 电子科技大学 | Plant covering information extracting method based on perfect polarization SAR images |
| CN104751183A (en) * | 2015-04-08 | 2015-07-01 | 西安电子科技大学 | Polarimetric SAR image classification method based on tensor MPCA |
| CN105930846A (en) * | 2016-04-05 | 2016-09-07 | 西安电子科技大学 | Neighborhood information and SVGDL (support vector guide dictionary learning)-based polarimetric SAR image classification method |
| CN107239757A (en) * | 2017-05-23 | 2017-10-10 | 西安电子科技大学 | A kind of polarization SAR silhouette target detection method based on depth ladder net |
| CN107292920A (en) * | 2017-06-15 | 2017-10-24 | 武汉大学 | A kind of multidate full-polarization SAR remote sensing imagery change detection method of joint classification |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108960295A (en) | 2018-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108960295B (en) | Multi-temporal fully-polarized SAR image feature extraction method and classification method | |
| Aghighi et al. | Machine learning regression techniques for the silage maize yield prediction using time-series images of Landsat 8 OLI | |
| Aksoy et al. | Automatic detection and segmentation of orchards using very high resolution imagery | |
| Loosvelt et al. | Impact of reducing polarimetric SAR input on the uncertainty of crop classifications based on the random forests algorithm | |
| Regniers et al. | Supervised classification of very high resolution optical images using wavelet-based textural features | |
| CN107103306B (en) | Winter wheat powdery mildew remote-sensing monitoring method based on wavelet analysis and support vector machines | |
| CN109325431B (en) | Method and device for detecting vegetation coverage in feeding path of grassland grazing sheep | |
| CN107229917A (en) | A kind of several remote sensing image general character well-marked target detection methods clustered based on iteration | |
| Gui et al. | A new method for soybean leaf disease detection based on modified salient regions | |
| CN108197650B (en) | Hyperspectral image extreme learning machine clustering method with local similarity maintained | |
| Chen et al. | Decision-level fusion of Sentinel-1 SAR and Landsat 8 OLI texture features for crop discrimination and classification: case of Masvingo, Zimbabwe | |
| CN110363236B (en) | Hyperspectral image extreme learning machine clustering method for embedding space-spectrum combined hypergraph | |
| Kenduiywo et al. | Higher order dynamic conditional random fields ensemble for crop type classification in radar images | |
| Ullagaddi et al. | Disease recognition in Mango crop using modified rotational kernel transform features | |
| CN110619263A (en) | Hyperspectral remote sensing image anomaly detection method based on low-rank joint collaborative representation | |
| Lu et al. | Comparison between several feature extraction/classification methods for mapping complicated agricultural land use patches using airborne hyperspectral data | |
| Chen et al. | Superpixel-based cropland classification of SAR image with statistical texture and polarization features | |
| Chen et al. | Delineating rice cropping activities from MODIS data using wavelet transform and artificial neural networks in the Lower Mekong countries | |
| CN107464247B (en) | Based on G0Distributed random gradient variational Bayesian SAR image segmentation method | |
| CN107610137A (en) | A kind of high-resolution remote sensing image optimal cut part method | |
| Ren et al. | Modified tensor distance-based multiview spectral embedding for PolSAR land cover classification | |
| Ma et al. | Land cover classification for polarimetric sar image using convolutional neural network and superpixel | |
| Kouskoulas et al. | The Bayesian hierarchical classifier (BHC) and its application to short vegetation using multifrequency polarimetric SAR | |
| Agmalaro et al. | Sentinel 1 classification for garlic land identification using Support Vector Machine | |
| CN114782835A (en) | Crop lodging area proportion detection method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20220815 Address after: 100190 No. 19 West North Fourth Ring Road, Haidian District, Beijing Applicant after: Research Institute of aerospace information innovation, Chinese Academy of Sciences Address before: No. 9 Dengzhuang South Road, Haidian District, Beijing 100094 Applicant before: Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences |
|
| TA01 | Transfer of patent application right |