CN114973012A - Cultivated land identification method and system of medium-low resolution remote sensing data - Google Patents

Cultivated land identification method and system of medium-low resolution remote sensing data Download PDF

Info

Publication number
CN114973012A
CN114973012A CN202210550678.1A CN202210550678A CN114973012A CN 114973012 A CN114973012 A CN 114973012A CN 202210550678 A CN202210550678 A CN 202210550678A CN 114973012 A CN114973012 A CN 114973012A
Authority
CN
China
Prior art keywords
image
evi
data
time sequence
pixel
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210550678.1A
Other languages
Chinese (zh)
Inventor
周玉洁
王谕锋
韩宇韬
吕琪菲
张至怡
陈爽
刘意
刘洋
余琴
冯敏铭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan Aerospace Shenkun Technology Co ltd
Original Assignee
Sichuan Aerospace Shenkun Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Sichuan Aerospace Shenkun Technology Co ltd filed Critical Sichuan Aerospace Shenkun Technology Co ltd
Priority to CN202210550678.1A priority Critical patent/CN114973012A/en
Publication of CN114973012A publication Critical patent/CN114973012A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/188—Vegetation
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
    • Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a farmland identification method and a farmland identification system of medium-low resolution remote sensing data, wherein the method comprises the following steps: obtaining EVI multi-time series data; filtering and reconstructing EVI multi-time sequence data by adopting a time sequence harmonic analysis method; performing dimensionality reduction and noise reduction on the EVI multi-time sequence data after filtering reconstruction by adopting minimum noise separation transformation to obtain an MNF image; performing pure pixel index calculation on the MNF image, and extracting a pure pixel image; selecting end member information of various ground features including cultivated land from the MNF image and the pure pixel image and constructing a ground feature sample library; and performing mixed pixel decomposition on the original image through a ground object sample library so as to obtain farmland image information in the original image. The method comprehensively utilizes the feature information of the ground features of the multi-time sequence of the ground features and the mixed pixel linear spectrum decomposition model, and avoids the phenomenon of wrong division of the ground features caused by difficulty in expressing the change of the ground features on the time spectrum, thereby improving the accuracy of the farmland information extraction.

Description

Cultivated land identification method and system of medium-low resolution remote sensing data
Technical Field
The invention relates to a method and a system for identifying cultivated land in medium-low resolution remote sensing image data.
Background
With the rapid development of satellite remote sensing technology, the remote sensing provides possibility for rapidly acquiring farmland information by the macroscopicity and real-time property of the remote sensing. Based on the medium-low resolution remote sensing data, the remote sensing technology is applied to the extraction of farmland information, and various classification algorithms are involved, and the remote sensing technology is mainly divided into two categories at present. The first is an unsupervised classification iterative self-organizing data analysis method (ISODATA) and a K-means mean classification algorithm, and the second is a maximum likelihood method, a Support Vector Machine (SVM), a Random Forest method (RF), a Decision Tree (DT) based on expert knowledge, a Neural Network (NN) and the like in the supervised classification.
The existing algorithm mainly has the following problems in the prior art that cultivated land information is extracted by using medium-low resolution remote sensing data: (1) the phase is single. The remote sensing image obtained in a certain time phase has the phenomenon that the same spectrum foreign matter is difficult to distinguish on the visible light spectrum. (2) For the areas with rich types of ground objects, the problem of mixing pixels is ubiquitous. The resolution of the medium-low resolution data is relatively low, the actual geographic space coverage is wide, and various types of ground objects may exist in a certain pixel, so that the accuracy of ground object extraction is influenced.
Disclosure of Invention
In view of the above, the invention provides a method and a system for recognizing cultivated land by using medium-low resolution remote sensing data, which couple time and space information and improve the accuracy of land feature extraction.
In order to solve the technical problems, the technical scheme of the invention is a cultivated land identification method adopting medium-low resolution remote sensing data, which comprises the following steps:
preprocessing a multi-phase satellite remote sensing original image to obtain EVI multi-time sequence data;
filtering and reconstructing EVI multi-time sequence data by adopting a time sequence harmonic analysis method;
performing dimensionality reduction and noise reduction on the EVI multi-time sequence data after filtering reconstruction by adopting minimum noise separation transformation to obtain an MNF image;
performing pure pixel index calculation on the MNF image, and extracting a pure pixel image;
combining the MNF image, the pure pixel image and the original image to select end member information of various ground features including cultivated land and construct a ground feature sample library;
and performing mixed pixel decomposition on the original image through a ground object sample library so as to obtain farmland image information in the original image.
As an improvement, the pretreatment comprises:
performing radiometric calibration, and converting the DN value of the image into earth surface radiance data;
performing atmospheric correction to obtain surface reflectivity data;
performing orthorectification on the image by using DEM data to generate a planar orthoimage;
and calculating the enhanced vegetation index in the image, and synthesizing enhanced vegetation index data of a plurality of time phases so as to obtain EVI multi-time sequence data.
As a further improvement, the method for filtering and reconstructing EVI multi-time series data by using time series harmonic analysis method includes:
generating a fitting curve, namely generating a least squares fitting curve by using discrete data in EVI multi-time series data;
removing deviation points, namely comparing the discrete data in the EVI multi-time series data with a fitted curve to be synthesized one by one, and removing points deviating from the fitted curve and exceeding a threshold value;
regenerating a fitting curve, namely regenerating the remaining discrete data into the fitting curve;
and repeatedly executing the steps of eliminating the deviation points and regenerating the fitted curve until the iteration times are reached or the fitted curve meeting the requirements is generated.
As another further improvement, the method for generating the fitting curve comprises:
let f (T) be a continuous time series signal with a period of 2T and satisfy the convergence theorem:
Figure BDA0003654917960000031
the Fourier expansion is as follows:
Figure BDA0003654917960000032
wherein the content of the first and second substances,
Figure BDA0003654917960000033
Figure BDA0003654917960000034
wherein j is the harmonic order, A 0 As harmonic remainder, A j Amplitude, ω, of the j-th harmonic j Is the frequency of the jth harmonic,
Figure BDA0003654917960000035
is the phase of the j-th harmonic; a is j 、b j Fourier coefficients for the jth harmonic of function f (t);
the Fourier coefficients are fitted by a least square method:
M T × M × J＝M T × Y
wherein J is a coefficient matrix, M is a Fourier matrix, M T Is the transpose of the fourier matrix.
As an improvement, the method for obtaining the MNF image by performing dimension reduction and noise reduction on the filtered and reconstructed EVI multi-time series data by using minimum noise separation transformation includes:
let W i (x) I is 1,2, …, n; n is the spatial dimension of the image; is provided with
W(x)＝M(x)+N(x)，
Wherein W (x) { W [ ((x) }) 1 (x),W 2 (x),…,W n (x) }; m (x) represents the signal in W (x), and N (x) is the noise in W (x).
Cov{W(x)}＝∑＝∑M+∑N
In the formula, Σ M represents a covariance matrix of a signal, and Σ N represents a covariance matrix of noise. The noise component of the i-th band is:
Var{N i (x)}/Var{W i (x)}。
as an improvement, the method for performing mixed pixel decomposition on an original image through a ground object sample library comprises the following steps:
respectively operating the ground feature sample models in the ground feature sample library on the basis of a linear spectrum mixed decomposition model for the pixels in the original image one by one to obtain the root mean square value of the pixels and each sample model;
selecting the minimum root mean square value of the pixel and each sample model, wherein the ground object represented by the ground object sample model with the minimum root mean square value is the ground object of the pixel;
and selecting all cultivated land pixels in the original image so as to generate cultivated land image information in the original image.
As an improvement, the linear spectrum mixed decomposition model is:
Figure BDA0003654917960000041
Figure BDA0003654917960000042
Figure BDA0003654917960000043
in the formula, N represents the number of end members; p represents any M-dimensional spectral vector in the image (M is the number of wave bands of the image); e denotes an mxn matrix, the columns denote end-member vectors, E ═ E 1 ,e 2 ,···,e n ](ii) a c represents a coefficient vector c ═ c 1 ,c 2 ,···,c n ) T ；c i Representing end-members e in a picture element i The ratio of (A) to (B); r represents an error term.
The invention also provides a cultivated land identification system of the medium-low resolution remote sensing data, which is characterized by comprising the following components:
the image preprocessing module is used for preprocessing the original images of the multi-stage satellite remote sensing to obtain EVI multi-time sequence data;
the EVI multi-time sequence data acquisition module is used for filtering and reconstructing the EVI multi-time sequence data by adopting a time sequence harmonic analysis method;
the dimensionality reduction and noise reduction module is used for carrying out dimensionality reduction and noise reduction on the EVI multi-time sequence data after filtering reconstruction by adopting minimum noise separation transformation to obtain an MNF image;
the pure pixel image acquisition module is used for calculating a pure pixel index of the MNF image and extracting a pure pixel image;
the surface feature sample library construction module is used for selecting end member information of various surface features including cultivated land from the MNF image and the pure pixel image and constructing a surface feature sample library;
and the cultivated land image information generation module is used for performing mixed pixel decomposition on the original image through the ground object sample library so as to obtain cultivated land image information in the original image.
As an improvement, the EVI multi-time series data acquisition module comprises:
the fitting curve generating module is used for generating a least square fitting curve from the discrete data;
and the deviation point removing module is used for comparing the discrete data in the EVI multi-time sequence data with the fitted curve one by one and removing the points which deviate from the fitted curve and exceed the threshold value.
As an improvement, the farmland image information generation module comprises:
the root mean square value solving module is used for respectively operating the ground feature sample models in the ground feature sample library on the basis of the linear spectrum mixed decomposition model for the pixels in the original image one by one to obtain the root mean square value of the pixels and each sample model;
the comparison module is used for selecting the minimum root mean square value of the pixel and each sample model, and the ground object represented by the ground object sample model with the minimum root mean square value is the ground object of the pixel;
and the cultivated land pixel selection module is used for selecting all cultivated land pixels in the original image so as to generate cultivated land image information in the original image.
The invention has the advantages that: the method comprehensively utilizes the feature spectral feature information of the features in the multi-time sequence and the mixed pixel linear spectral decomposition model based on the data with medium and low resolution, avoids the phenomenon of mistaken division of the features caused by difficulty in expressing the change of the features on the time spectrum based on the traditional spectral feature classification method, and further improves the accuracy of the farmland information extraction.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a schematic diagram of the structure of the present invention.
Detailed Description
In order that those skilled in the art will better understand the technical solutions of the present invention, the present invention will be further described in detail with reference to the following embodiments.
Example 1
As shown in FIG. 1, the invention provides a farmland identification method of medium-low resolution remote sensing data, which comprises the following steps:
s1, preprocessing the original images of the multi-stage satellite remote sensing to obtain EVI multi-time sequence data. The Enhanced Vegetation Index (EVI) can simultaneously reduce the influence from atmospheric noise and soil noise and stably reflect the Vegetation condition of a tested region. The range setting of red light and near-infrared detection wave band is narrower, has not only improved the ability of being surveyed rare sparse planting, has reduced the influence of steam moreover, simultaneously, has introduced the blue light wave band and has corrected atmospheric aerosol's scattering and soil background.
And S2, filtering and reconstructing the EVI multi-time-series data by adopting a time-series harmonic analysis method. Due to the influences of factors such as solar altitude, observation angle, water vapor, aerosol and cloud, the time sequence data of various places are irregular in change, the time sequence data have high volatility, the change trend and the characteristic of the phenological information in a curve year are not obvious, and the trend analysis and the information extraction cannot be effectively carried out on the time dimension. The enhanced vegetation index EVI Time Series curve can be reconstructed by using a Time Series Harmonic Analysis method (Harmonic Analysis of Time Series, HANTS), so that the situation is improved.
And S3, performing dimensionality reduction and noise reduction on the EVI multi-time sequence data after filtering reconstruction by adopting minimum noise separation transformation to obtain an MNF image. Minimum Noise separation transform (MNF) is a spatial map-bit transform method that reduces the Noise information of data. The essence of the method is two times of principal component method transformation, and compared with a principal component analysis method, the method can better perform dimensionality reduction processing on multispectral remote sensing image data, eliminate correlation among wave bands and compress data quantity.
S4, pure pixel index calculation is carried out on the MNF image, and a pure pixel image is extracted; the pure picture element refers to a picture element only containing one type of geological spectral information. Based on the MNF image, the pure pixel index image is obtained, and the accuracy of various land samples such as cultivated land can be improved. First, the data is subjected to noise and dimension reduction processing by MNF transformation. And if the unit vector is taken as a reference, projecting all pixel points of the feature space to the pixel points. Then, pure pixels are distributed on two sides of the unit; the middle part is provided with mixed pixels. Thus, when an image is projected on N random unit vectors, the number of times each pel is projected to an endpoint is PPI. Based on PPI images, pure pixels can be rapidly positioned, and the probability of selecting pure sampling points can be improved.
S5, selecting end member information of various land features including cultivated land from the MNF image and the pure pixel image and constructing a land feature sample library; the ground feature sample library is constructed by selecting samples of different ground features on an image in a man-machine interaction mode by using a pure pixel image (PPI image), a minimum noise separation image (MNF image) and an N-dimensional visualization tool.
S6, mixed pixel decomposition is carried out on the original image through the ground feature sample library, and therefore cultivated land image information in the original image is obtained. The mixed image element refers to an image element containing spectral information of various ground objects. And selecting and processing the pixels containing farmland spectral information by decomposing the mixed pixels to obtain a farmland image in the original remote sensing image.
Example 2
Based on embodiment 1, the preprocessing in step S1 includes:
s11, performing radiometric calibration, and converting the DN value of the image into earth surface radiance data;
s12, performing atmospheric correction to obtain earth surface reflectivity data;
s13, performing orthorectification on the image by using DEM data to generate a planar orthorectified image;
s14 calculates an enhanced vegetation index in the image, and synthesizes enhanced vegetation index data for a plurality of time phases to obtain EVI multi-time series data.
In the case of the example 3, the following examples are given,
based on embodiment 1, step S2 employs a time-series harmonic analysis method to perform filtering reconstruction on EVI multi-time-series data. The main principle of harmonic analysis is to transform a time-series signal f (t) of an element from time space to frequency space by discrete fourier transform. For each frequency component, a harmonic signal corresponds to it in the time domain space. HANTS is to reconstruct time series data by decomposing the time series data into several harmonic curves of different frequencies, and then selecting and superposing the curves capable of reflecting the characteristics of the time series. The specific method comprises the following steps:
s21 generating a fitting curve, namely generating a least squares fitting curve by using discrete data in EVI multi-time series data; the method for generating the fitting curve comprises the following steps:
let f (T) be a continuous time series signal with a period of 2T and satisfy the convergence theorem:
Figure BDA0003654917960000081
the Fourier expansion is as follows:
Figure BDA0003654917960000082
wherein the content of the first and second substances,
Figure BDA0003654917960000083
Figure BDA0003654917960000084
wherein j is the harmonic order, A 0 As harmonic remainder, A j Amplitude, ω, of the j-th harmonic j Is the frequency of the jth harmonic,
Figure BDA0003654917960000085
is the phase of the j-th harmonic; a is j 、b j Fourier coefficients for the jth harmonic of function f (t);
the Fourier coefficients are fitted by a least square method:
M T × M × J＝M T × Y
wherein J is a coefficient matrix, M is a Fourier matrix, M T Is the transpose of the fourier matrix.
And S22, removing deviation points, comparing the discrete data in the EVI multi-time sequence data with a fitted curve to be synthesized one by one, and removing the points of which the deviation of the fitted curve exceeds a threshold value.
S23 regenerating a fitting curve, namely regenerating the remaining discrete data into the fitting curve; the method for generating the fitting curve in this step is the same as that in step S21, and is not described herein again.
S24, the step of eliminating the deviation points and the step of regenerating the fitting curve are repeatedly executed until the iteration number is reached or the fitting curve meeting the requirement is generated.
It should be noted that in the process of filtering and reconstructing the time series data by using the hatts, 5 key parameters are very important, and the settings need to be continuously adjusted so as to obtain a better curve fitting effect.
(1) Valid numerical range (Valid data): the effective value range is used for eliminating data values without physical significance, data outside the effective value range are invalid, and the data are directly eliminated without participating in the operation process of curve fitting.
(2) Frequency (Number of frequencies): the frequency count directly determines the harmonic order into which the time series data is decomposed. The frequency i (i equals 1,2, …) results in a harmonic curve with a period of 1/i time series length (i equals 0, which results in an average over the entire time series length).
(3) High-low inhibition markers (Direction outpiers): valid value points above or below a certain limit of the fitted curve should be reconstructed.
(4) Fit Error (Fit Error Tolerance, FET): if the absolute distance between the deviation point and the curve exceeds a certain threshold value, the deviation points are removed, the rest points can reconstruct a fitting curve, the process is repeated until the distances between all the rest points and the curve are within the fitting error range, and the rest points are the final screening results. It should be noted that if the fitting error value is too large, the number of points to be removed is too small, and the cloud influence cannot be completely removed; the fitting error value is too small, and the removed points are too many, so that the authenticity of curve fitting cannot be ensured.
(5) Number of limit points (Degrid of Over decision, DOD): on the premise of conforming to the fitting error value, if too many points are removed, the remained original value points are too few, and the curve cannot be reconstructed, so that the fitting curve and the original curve have larger errors. Generally, the minimum number of original value points needs to be kept as (2 × frequency-1), and in practical operation, the DOD is often larger than the critical value.
Example 4
Based on embodiment 1, the method for obtaining the MNF image by performing the dimension reduction and noise reduction on the filtered and reconstructed EVI multi-time series data by using the minimum noise separation transform in step S3 includes:
let W i (x) I is 1,2, …, n; n is the spatial dimension of the image; where W (x) ═ M (x) + N (x),
wherein W (x) { W [ ((x) }) 1 (x),W 2 (x),…,W n (x) }; m (x) represents the signal in W (x), and N (x) is the noise in W (x).
Cov{W(x)}＝∑＝∑M+∑N
In the formula, Σ M represents a covariance matrix of a signal, and Σ N represents a covariance matrix of noise. The noise component of the i-th band is:
Var{N i (x)}/Var{W i (x)}。
example 5
Based on embodiment 1, the method for performing mixed pixel decomposition on the original image through the surface feature sample library in step S6 to obtain arable land image information in the original image includes:
s61, respectively operating the ground feature sample models in the ground feature sample library on the pixels in the original image one by one based on the linear spectrum mixed decomposition model to obtain the root mean square value of the pixels and each sample model;
the linear spectrum mixed decomposition model is as follows:
Figure BDA0003654917960000101
Figure BDA0003654917960000102
Figure BDA0003654917960000103
in the formula, N represents the number of end members; p represents any M-dimensional spectral vector in the image (M is the number of wave bands of the image); e denotes an mxn matrix, the columns denote end-member vectors, E ═ E 1 ,e 2 ,···,e n ](ii) a c represents a coefficient vector c ═ c 1 ,c 2 ,···,c n ) T ；c i Representing end-members e in a picture element i The ratio of (A) to (B); r represents an error term.
S62, selecting the minimum root mean square value of the pixel and each sample model, wherein the ground object represented by the ground object sample model with the minimum root mean square value is the ground object of the pixel;
s63, all cultivated land pixels in the original image are selected to generate cultivated land image information in the original image.
For example, for the pixels located at (100 ), all the surface feature sample models in the surface library sample library are run respectively (if there are 1000). If the result is RMSE10< RMSE4< RMSE98<. > < RMSE63, according to the principle of RMSE minimum, the end member type corresponding to the 10 th model is selected as the land class of the pixel (100 ), and the end member type corresponding to the 10 th end member model is cultivated land, then the (100 ) is cultivated land. And so on until each pixel in the image is decomposed. And finally, integrating the arable land pixels.
Example 6
As shown in FIG. 2, the present invention further provides a cultivated land identification system of medium-low resolution remote sensing data, comprising:
the image preprocessing module is used for preprocessing the original images of the multi-stage satellite remote sensing to obtain EVI multi-time sequence data;
the EVI multi-time sequence data acquisition module is used for filtering and reconstructing the EVI multi-time sequence data by adopting a time sequence harmonic analysis method;
the dimensionality reduction and noise reduction module is used for carrying out dimensionality reduction and noise reduction on the EVI multi-time sequence data after filtering reconstruction by adopting minimum noise separation transformation to obtain an MNF image;
the pure pixel image acquisition module is used for calculating a pure pixel index of the MNF image and extracting a pure pixel image;
the surface feature sample library construction module is used for selecting end member information of various surface features including cultivated land from the MNF image and the pure pixel image and constructing a surface feature sample library;
and the cultivated land image information generation module is used for performing mixed pixel decomposition on the original image through the ground object sample library so as to obtain cultivated land image information in the original image.
Example 7
Based on embodiment 6, the EVI multi-time series data acquisition module specifically includes:
the fitting curve generating module is used for generating a least square fitting curve from the discrete data;
and the deviation point removing module is used for comparing the discrete data in the EVI multi-time sequence data with the fitted curve one by one and removing the points which deviate from the fitted curve and exceed the threshold value.
Example 8
Based on embodiment 6, the arable land image information generation module includes:
the root mean square value solving module is used for respectively operating the ground feature sample models in the ground feature sample library on the basis of the linear spectrum mixed decomposition model for the pixels in the original image one by one to obtain the root mean square value of the pixels and each sample model;
the comparison module is used for selecting the minimum root mean square value of the pixel and each sample model, and the ground object represented by the ground object sample model with the minimum root mean square value is the ground object of the pixel;
and the cultivated land pixel selection module is used for selecting all cultivated land pixels in the original image so as to generate cultivated land image information in the original image.
The invention provides a method for extracting farmland information based on medium-low resolution remote sensing images by coupling time and space information, which comprises the following steps: on a time scale, enhanced vegetation indexes of a plurality of months can be combined, the phenological information of cultivated land is comprehensively utilized, a multi-time series ground feature curve is constructed, and cultivated land information can be optimally extracted from a single time-phase remote sensing image; on the basis of spatial spectrum information, considering that the area size of the medium-low resolution remote sensing data is large, the number of mixed pixels is large, a mixed pixel decomposition method is applied, a plurality of different ground feature model combinations are operated in one pixel, and finally the optimal solution is selected as the ground feature type of the pixel, so that the precision of farmland extraction is improved. The method integrates the time scale and the space scale, and can effectively improve the accuracy of farmland information extraction in the medium-low resolution data.
The above is only a preferred embodiment of the present invention, and it should be noted that the above preferred embodiment should not be considered as limiting the present invention, and the protection scope of the present invention should be subject to the scope defined by the claims. It will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the spirit and scope of the invention, and these modifications and adaptations should be considered within the scope of the invention.

Claims (10)

1. A farmland identification method of medium-low resolution remote sensing data is characterized by comprising the following steps:
preprocessing a multi-phase satellite remote sensing original image to obtain EVI multi-time sequence data;
filtering and reconstructing EVI multi-time sequence data by adopting a time sequence harmonic analysis method;
performing dimensionality reduction and noise reduction on the EVI multi-time sequence data after filtering reconstruction by adopting minimum noise separation transformation to obtain an MNF image;
performing pure pixel index calculation on the MNF image, and extracting a pure pixel image;
selecting end member information of various ground features including cultivated land from the MNF image and the pure pixel image and constructing a ground feature sample library;
and performing mixed pixel decomposition on the original image through a ground object sample library so as to obtain farmland image information in the original image.
2. The method for farmland identification of medium-low resolution remote sensing data according to claim 1, characterized in that the preprocessing comprises:
performing radiometric calibration, and converting the DN value of the image into earth surface radiance data;
performing atmospheric correction to obtain surface reflectivity data;
performing orthorectification on the image by using DEM data to generate a planar orthoimage;
and calculating the enhanced vegetation index in the image, and synthesizing enhanced vegetation index data of a plurality of time phases so as to obtain EVI multi-time sequence data.
3. The method for identifying cultivated land by using medium-low resolution remote sensing data according to claim 1, wherein the method for filtering and reconstructing EVI multi-time sequence data by using time sequence harmonic analysis comprises the following steps:
generating a fitting curve, namely generating a least squares fitting curve by using discrete data in EVI multi-time series data;
removing deviation points, namely comparing discrete data in the EVI multi-time sequence data with a fitted curve to be synthesized one by one, and removing points which deviate from the fitted curve and exceed a threshold value;
regenerating a fitted curve, namely regenerating the fitted curve from the residual discrete data;
and repeatedly executing the steps of eliminating the deviation points and regenerating the fitted curve until the iteration times are reached or the fitted curve meeting the requirements is generated.
4. The method for identifying the farmland based on the medium-low resolution remote sensing data according to claim 3, characterized in that the method for generating the fitting curve is as follows:
let f (T) be a continuous time series signal with a period of 2T and satisfy the convergence theorem:
Figure FDA0003654917950000021
the Fourier expansion is as follows:
Figure FDA0003654917950000022
wherein the content of the first and second substances,
Figure FDA0003654917950000023
Figure FDA0003654917950000024
wherein j is the harmonic order, A 0 As harmonic remainder, A j Amplitude, ω, of the j-th harmonic j Is the frequency of the jth harmonic,
Figure FDA0003654917950000025
is the phase of the j-th harmonic; a is j 、b j Fourier coefficients for the jth harmonic of function f (t);
the Fourier coefficients are fitted by a least square method:
M T ×M×J＝M T ×Y
wherein J is a coefficient matrix, M is a Fourier matrix, M T Is the transpose of the fourier matrix.
5. The method for identifying cultivated land by using medium-low resolution remote sensing data according to claim 1, wherein the method for obtaining MNF images by performing dimension reduction and noise reduction on EVI multi-time series data after filtering reconstruction by adopting minimum noise separation transformation comprises the following steps:
let W i (x) I is 1,2, …, n; n is the spatial dimension of the image; is provided with
W(x)＝M(x)+N(x)，
Wherein W (x) { W [ ((x) }) 1 (x),W 2 (x),…,W n (x) }; m (x) represents the signal in W (x), and N (x) is the noise in W (x).
Cov{W(x)}＝∑＝∑M+∑N
In the formula, Σ M represents a covariance matrix of a signal, and Σ N represents a covariance matrix of noise. The noise component of the i-th band is:
Var{N i (x)}/Var{W i (x)}。
6. the method for identifying cultivated land by using medium-low resolution remote sensing data according to claim 1, wherein the method for performing mixed pixel decomposition on the original image by using the surface feature sample library comprises the following steps:
respectively operating the ground feature sample models in the ground feature sample library on the basis of a linear spectrum mixed decomposition model for the pixels in the original image one by one to obtain the root mean square value of the pixels and each sample model;
selecting the minimum root mean square value of the pixel and each sample model, wherein the ground object represented by the ground object sample model with the minimum root mean square value is the ground object of the pixel;
and selecting all cultivated land pixels in the original image after model operation so as to generate cultivated land image information in the original image.
7. The method for identifying cultivated land by using medium-low resolution remote sensing data according to claim 6, wherein the linear spectrum mixed decomposition model is as follows:
Figure FDA0003654917950000031
Figure FDA0003654917950000032
Figure FDA0003654917950000033
in the formula, N represents the number of end members; p represents any M-dimensional spectral vector in the image (M is the number of wave bands of the image); e represents an M × N matrix, and columns represent end membersVector, E ═ E 1 ,e 2 ,···,e n ](ii) a c represents a coefficient vector c ═ c 1 ,c 2 ,···,c n ) T ；c i Representing end-members e in a picture element i The ratio of (A) to (B); r represents an error term.
8. A farmland identification system of medium-low resolution remote sensing data is characterized by comprising:
the image preprocessing module is used for preprocessing the original images of the multi-stage satellite remote sensing to obtain EVI multi-time sequence data;
the EVI multi-time sequence data acquisition module is used for filtering and reconstructing the EVI multi-time sequence data by adopting a time sequence harmonic analysis method;
the dimensionality reduction and noise reduction module is used for carrying out dimensionality reduction and noise reduction on the EVI multi-time sequence data after filtering reconstruction by adopting minimum noise separation transformation to obtain an MNF image;
the pure pixel image acquisition module is used for calculating a pure pixel index of the MNF image and extracting a pure pixel image;
the surface feature sample library construction module is used for selecting end member information of various surface features including cultivated land from the MNF image and the pure pixel image and constructing a surface feature sample library;
and the cultivated land image information generation module is used for performing mixed pixel decomposition on the original image through the ground object sample library so as to obtain cultivated land image information in the original image.
9. The tilled land identification system based on medium-low resolution remote sensing data according to claim 8, wherein the EVI multi-time series data acquisition module comprises:
the fitting curve generating module is used for generating a least square fitting curve from the discrete data;
and the deviation point removing module is used for comparing the discrete data in the EVI multi-time sequence data with the fitted curve one by one and removing the points which deviate from the fitted curve and exceed the threshold value.
10. The farmland identification system of the medium-low resolution remote sensing data according to claim 8, characterized in that the farmland image information generation module comprises:
the root mean square value solving module is used for respectively operating the ground feature sample models in the ground feature sample library on the basis of the linear spectrum mixed decomposition model for the pixels in the original image one by one to obtain the root mean square value of the pixels and each sample model;
the comparison module is used for selecting the minimum root mean square value of the pixel and each sample model, and the ground object represented by the ground object sample model with the minimum root mean square value is the ground object of the pixel;
and the cultivated land pixel selection module is used for selecting all cultivated land pixels in the original image so as to generate cultivated land image information in the original image.
CN202210550678.1A 2022-05-20 2022-05-20 Cultivated land identification method and system of medium-low resolution remote sensing data Pending CN114973012A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210550678.1A CN114973012A (en) 2022-05-20 2022-05-20 Cultivated land identification method and system of medium-low resolution remote sensing data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210550678.1A CN114973012A (en) 2022-05-20 2022-05-20 Cultivated land identification method and system of medium-low resolution remote sensing data

Publications (1)

Publication Number Publication Date
CN114973012A true CN114973012A (en) 2022-08-30

Family

ID=82985421

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210550678.1A Pending CN114973012A (en) 2022-05-20 2022-05-20 Cultivated land identification method and system of medium-low resolution remote sensing data

Country Status (1)

Country Link
CN (1) CN114973012A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115983503A (en) * 2023-03-18 2023-04-18 杭州领见数字农业科技有限公司 Crop maturity prediction method, equipment and storage medium

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115983503A (en) * 2023-03-18 2023-04-18 杭州领见数字农业科技有限公司 Crop maturity prediction method, equipment and storage medium

Similar Documents

Publication Publication Date Title
Lu et al. Removal of noise by wavelet method to generate high quality temporal data of terrestrial MODIS products
Roberts et al. Assessment of image fusion procedures using entropy, image quality, and multispectral classification
US6804400B1 (en) Adaptive hyperspectral data compression
Regniers et al. Supervised classification of very high resolution optical images using wavelet-based textural features
Mundt et al. Partial unmixing of hyperspectral imagery: theory and methods
CN114926748A (en) Soybean remote sensing identification method combining Sentinel-1/2 microwave and optical multispectral images
CN114973012A (en) Cultivated land identification method and system of medium-low resolution remote sensing data
CN113421198B (en) Hyperspectral image denoising method based on subspace non-local low-rank tensor decomposition
Cresson et al. Comparison of convolutional neural networks for cloudy optical images reconstruction from single or multitemporal joint SAR and optical images
Muhammad et al. Evaluation of wavelet transform algorithms for multi-resolution image fusion
Cerra et al. Cloud removal in image time series through sparse reconstruction from random measurements
Mani A survey of multi sensor satellite image fusion techniques
De Grandi et al. Spatial wavelet statistics of SAR backscatter for characterizing degraded forest: A case study from Cameroon
Alipour Fard et al. Classifier fusion of high-resolution optical and synthetic aperture radar (SAR) satellite imagery for classification in urban area
Qian et al. Effect of lossy vector quantization hyperspectral data compression on retrieval of red-edge indices
Meneghini An Evaluation of Sentinel-1 and Sentinel-2 for Land Cover Classification
Li et al. Improving the accuracy of linear pixel unmixing via appropriate endmember dimensionality reduction
Borghys et al. Change detection in urban scenes by fusion of SAR and hyperspectral data
CN117689579B (en) SAR auxiliary remote sensing image thick cloud removal method with progressive double decoupling
Das et al. A Comparative Analysis of Image Fusion Techniques for Remote Sensed Images.
Arai Combined Non-Parametric and Parametric Classification Method Depending on Normality of PDF of Training Samples
Zhang et al. Hyperspectral Data Fusion Using Multidimensional Information
Robila et al. Exploring independent component analysis for remote sensing
Van Trung et al. Fusion of ALOS PALSAR and ASTER data for landcover classification at Tonle Sap floodplain, Cambodia
De Carvalho et al. Multiscale feature extraction from images using wavelets

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