CN114120027A - Phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data - Google Patents
Phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data Download PDFInfo
- Publication number
- CN114120027A CN114120027A CN202111233559.5A CN202111233559A CN114120027A CN 114120027 A CN114120027 A CN 114120027A CN 202111233559 A CN202111233559 A CN 202111233559A CN 114120027 A CN114120027 A CN 114120027A
- Authority
- CN
- China
- Prior art keywords
- image sequence
- historical
- vegetation sample
- sample image
- historical vegetation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a phenological extraction and land cover classification method based on MODIS long-time sequence data, which is characterized in that based on a historical vegetation sample image sequence, an MRT tool is adopted to update the historical vegetation sample image sequence corresponding to a target area, then GlobalLand30 is applied to obtain images of all target object types in the historical vegetation sample image sequence corresponding to the target area as all samples, a vertex component analysis method is applied to screen all the samples, training is combined with a classifier to obtain a target classifier, then the target classifier is applied to obtain the target object types respectively corresponding to all pixel areas in the historical vegetation sample image sequence corresponding to the target area, finally the distribution of all the phenological markers in the target area space and the evolution rule of the phenological markers along with seasonal parameters on the time sequence are obtained, the whole technical scheme is based on rich time characteristics, the land utilization/land cover types can be effectively distinguished, and then the efficient and accurate earth surface coverage classification is obtained.
Description
Technical Field
The invention relates to a phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data, and belongs to the technical field of earth surface coverage analysis.
Background
With the development of society and the progress of cities, various ecological environmental problems ensue, and the ecological environmental problems directly or indirectly influence the climate change of urban vegetation. The land cover classification is carried out according to the phenological information, so that the classification precision can be improved, and the method has certain significance for researching land utilization. In the current research on regions, most of the existing time series classification methods using a single classifier are based on a single classifier, and the change of the representation time series can deteriorate the discrimination, the long-time series data classification is difficult, and is easily influenced by signal pollution, and the data classification is not comprehensive and accurate.
Disclosure of Invention
The invention aims to provide a phenological extraction and land cover classification method based on MODIS long-time sequence data, which can more effectively distinguish land utilization/land cover types based on abundant time characteristics so as to obtain efficient and accurate land cover classification.
The invention adopts the following technical scheme for solving the technical problems: the invention designs a phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data, which comprises the following steps:
a, obtaining an MODIS global vegetation index image sequence which corresponds to a target area and has a preset historical time range and contains NDVI and EVI, taking the obtained MODIS global vegetation index image sequence as a historical vegetation sample image sequence corresponding to the target area, and then entering the step B;
b, adopting an MRT tool to perform batch processing on the historical vegetation sample image sequence corresponding to the target area, converting the Sinusoid projection in the historical vegetation sample image sequence into a preset target space coordinate system, updating the historical vegetation sample image sequence corresponding to the target area, and entering the step C;
step C, obtaining images of each target object type in the historical vegetation sample image sequence corresponding to the target area by using GlobalLand30 as each sample, and then entering step D;
d, screening all samples by using a vertex component analysis method to obtain each training sample, training the training samples by using the training samples as input and the target object types corresponding to the training samples as output, training each preset classifier to be trained to obtain a trained classifier corresponding to the highest accuracy as a target classifier, and then entering the step E;
e, classifying each pixel region in the historical vegetation sample image sequence corresponding to the target region by applying a target classifier to obtain target object types respectively corresponding to each pixel region in the historical vegetation sample image sequence corresponding to the target region, and then entering the step F;
f, aiming at the historical vegetation sample image sequence corresponding to the target area, performing peak elimination by using STL decomposition, extracting seasonal parameters in the historical vegetation sample image sequence, performing fitting filtering on the time of the historical vegetation sample image sequence by using a Savitzky-Golay filter, updating the historical vegetation sample image sequence, and then entering the step G;
and G, marking the phenological information of the historical vegetation sample image sequence by using a TIMEAT time sequence analysis method based on the target object type corresponding to each pixel region in the historical vegetation sample image sequence corresponding to the target region, and obtaining the distribution of each phenological mark on the target region space and the evolution rule of each phenological mark along with seasonal parameters on a time sequence.
As a preferred technical scheme of the invention: in the step B, the MRT tool is used to batch process the image sequence of the historical vegetation sample corresponding to the target area, the sinussoid projection in the image sequence of the historical vegetation sample is converted into WGS _1984_ UTM _ Zone _50N, and the image sequence of the historical vegetation sample corresponding to the target area is updated.
As a preferred technical scheme of the invention: and in the step C, obtaining images of target object types of cultivated land, forest, grassland, wetland, water body and artificial surface in the historical vegetation sample image sequence corresponding to the target area by using GlobalLand30 as each sample.
As a preferred technical scheme of the invention: and D, presetting each classifier to be trained to comprise a rotary forest classifier, a Bagging classifier and a random forest classifier.
As a preferred embodiment of the present invention, the step F includes the following steps F1 to F2:
step F1, aiming at a historical vegetation sample image sequence corresponding to a target area, performing peak removal by using STL decomposition, and extracting seasonal parameters in the historical vegetation sample image sequence, namely aiming at the historical vegetation sample image sequence, decomposing the historical vegetation sample image sequence into seasons, trends and remainders as follows;
Yt＝St+Tt+et (1)
where t is 1, …, n, n represents the total number of time nodes corresponding to the image sequence of the historical vegetation sample, and Y represents the total number of time nodes corresponding to the image sequence of the historical vegetation sampletIs the data of the corresponding time node t in the image sequence of the historical vegetation sample StRepresenting seasonal components, T, corresponding to the decomposition of the corresponding time node T in the image sequence of the historical vegetation samplestRepresenting trend components, e, corresponding to the decomposition of the corresponding time node t in the image sequence of the historical vegetation sampletRepresenting a remainder component corresponding to the decomposition of the corresponding time node t in the historical vegetation sample image sequence;
wherein, m seasonal breakpoints in the image sequence are preset according to the historical vegetation samples
And define
Based on j being 1, …, m, StThe corresponding harmonic model with the K harmonic term is represented as follows:
wherein m represents the number of seasonal breakpoints in the preset historical vegetation sample image sequence, aj,kRepresenting the amplitude, delta, of the image sequence of the historical vegetation sample corresponding to the jth seasonal breakpoint and the kth harmonic analysisj,kRepresenting the image sequence of the historical vegetation sample corresponding to the jth seasonal breakThe point and the phase under the kth harmonic analysis, wherein f represents the frequency of each historical vegetation sample image segment of each seasonal breakpoint corresponding to the historical vegetation sample image sequence;
according to TtAnd the slope beta of each historical vegetation sample image segment of each seasonal breakpoint corresponding to the historical vegetation sample image sequenceiIntercept αiObtaining TtIs represented as follows:
Tt＝αi+βit (3)
where i is 1, …, m +1, i.e. m +1 image segments of the historical vegetation sample, βiRepresenting the slope, α, of the image segment of the ith historical vegetation sampleiRepresenting an intercept representing an image segment of the ith historical vegetation sample;
step F2. is based on the decomposition of the historical vegetation sample image sequence with respect to season, trend, and residue, applying Savitzky-Golay filter, in combination with the sliding of the window in the historical vegetation sample image sequence, as follows:
the average value p of the linear combination of the window neighborhood values is used for each data y in the windowqThe fitting filtering to the image sequence of the historical vegetation sample is realized, wherein Q is 1, …, Q represents the number of data in the window, yqRepresenting the q-th data within the window, cgIs represented by cg＝1/(2n+1)。
Compared with the prior art, the phenological extraction and surface coverage classification method based on MODIS long-time sequence data has the following technical effects:
(1) the invention relates to a phenological extraction and ground surface coverage classification method based on MODIS long-time sequence data, which comprises the steps of firstly obtaining a historical vegetation sample image sequence, then adopting an MRT tool to carry out batch processing on the historical vegetation sample image sequence corresponding to a target area, updating the historical vegetation sample image sequence corresponding to the target area, then applying GlobalLand30 to obtain an image of each target object type in the historical vegetation sample image sequence corresponding to the target area as each sample, screening all the samples by applying a vertex component analysis method, combining with classifier training to obtain a target classifier, then applying the target classifier to carry out classification to obtain target object types respectively corresponding to each pixel area in the historical vegetation sample image sequence corresponding to the target area, combining with extraction of seasonal parameters in the historical vegetation sample image sequence, and fitting and filtering time of the historical vegetation sample image sequence, and updating the historical vegetation sample image sequence, and finally obtaining the distribution of each phenological mark on the target area space and the evolution rule of each phenological mark along with seasonal parameters on the time sequence.
Drawings
FIG. 1 is a schematic diagram of the structure of a phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data according to the present invention;
FIG. 2 is a pictorial illustration of target object types obtained using GlobalLand30 in an embodiment;
FIG. 3 is a schematic representation of the results of batch processing of an image sequence of a sample of historical vegetation by an MRT tool according to an embodiment;
Detailed Description
The following description will explain embodiments of the present invention in further detail with reference to the accompanying drawings.
The invention designs a phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data, which is based on the figure 1 and concretely executes the following steps A to G in practical application.
And A, obtaining an MODIS global vegetation index image sequence which corresponds to a target area and has a preset historical time range and contains NDVI and EVI, taking the obtained MODIS global vegetation index image sequence as a historical vegetation sample image sequence corresponding to the target area, and then entering the step B.
The MODIS data is improved in data performance and quality in many ways compared to NOAA/AVHRR. The spatial resolution is improved to 250 m; the time resolution is also improved; the system has 36 spectrum observation channels, so that the spectrum resolution is greatly improved, and the observation capability of the earth complex system and the identification capability of the earth surface type are greatly enhanced; and MODIS data has the advantage of being free and simple to accept. The advantages enable MODIS data to greatly improve the quality and timeliness of remote sensing data.
The MODIS vegetation index is generated at 16-day intervals and multiple spatial resolutions, providing a spatiotemporal comparison of the composite properties of vegetation canopy greenness, leaf area, chlorophyll and canopy structure; wherein the two vegetation indexes are derived from the atmospheric correction reflectivity of infrared, near infrared and blue wave bands; the AVHRR NDVI time series record for NOAA provides a continuous Normalized Differential Vegetation Index (NDVI) for historical and climatic applications and an Enhanced Vegetation Index (EVI) that minimizes canopy soil variation and improves its sensitivity under dense vegetation conditions. Both products more effectively characterize vegetation status and processes on a global scale.
And B, adopting an MRT tool to perform batch processing on the historical vegetation sample image sequence corresponding to the target area, converting the Sinusoid projection in the historical vegetation sample image sequence into a WGS _1984_ UTM _ Zone _50N space coordinate system, updating the historical vegetation sample image sequence corresponding to the target area, as shown in FIG. 3, and then entering the step C. In a particular practical application, this is considered as a standard projection and coordinate system. A bat document was developed for stitching, reprojection, resampling, and converting MODIS products to GeoTIFF documents.
And C, obtaining images of target object types of farmland, forests, grasslands, wetlands, water bodies and artificial surfaces in the historical vegetation sample image sequence corresponding to the target area by using the GlobalLand30, using the images as the samples as shown in figure 2, and then entering the step D.
And D, screening all samples by using a Vertex Component Analysis (VCA) method to obtain each training sample, taking the training sample as input, taking the type of a target object corresponding to the training sample as output, training each classifier to be trained, which is preset to comprise a rotating Forest classifier (RoF), a Bagging classifier and a Random Forest classifier (RF), obtaining a classifier after training corresponding to the highest accuracy as a target classifier, and entering the step E.
The goal of ensemble learning is to combine the results of many vulnerable learners with a comprehensive predictor, aiming to improve classification accuracy and stability. Multiple classifiers can be designed independently without any interaction, their outputs combined according to a given strategy. Generating such an efficient set is the core of this paradigm, and the combination of base learners, vulnerable learners, and combination strategies often differ between the various methods of integrated learning. The Bagging classifier, the Random Forest classifier (RF) and the rotating Forest classifier (RoF) are respectively dedicated to training data sampling, feature selection and feature extraction to generate a relevant classifier set.
The Bagging classifier is based on repeatable sampling (Bootsta sampling), wherein during each training, the training set is subjected to random sampling with the back being put according to the principle of uniform distribution to obtain a new training set (also called a bag (bag)), and each new training set contains 63.2% of samples of the original data set on average, and the samples in the original data set may appear in the new training set one or more times or may not appear. After repeating T times, training T base classifiers by using the T sample sets. And finally, integrating the results obtained by the T base classifiers by adopting a majority voting method.
The basic idea of the RoF algorithm is to randomly extract a feature data set of original data, divide the feature data set of the original data into a plurality of subsets, perform feature transformation on each subset, and merge the obtained transformation components according to the original sequence of the feature subsets. The rotation forest uses a feature extraction algorithm to generate a sparse rotation matrix, so that the original image is projected to different coordinate systems, and the constructed base classifier has strong difference. Because the rotary Senri can construct a more different base classifier, the performance of the rotary Senri is often better than that of traditional ensemble learning classification algorithms such as Bagging and Ada Boost.
The RF algorithm can be regarded as a combination of Bagging and RS, and is a combination of a series of classifiers to make a decision, and it is expected to obtain a most "fair" ensemble learning method. Constructing each classifier requires randomly extracting a part of samples from an original data set as a sample subspace, then randomly selecting a new feature subspace from the sample subspace, establishing a decision tree in the new space as a classifier, and finally obtaining a final decision through a voting method. Compared with the traditional decision tree, the RF has stronger generalization capability and better classification effect.
Here, to avoid bias evaluation from training data initialization, VCA is applied in step D above to select more reliable samples for each classifier. In each run, 100 labeled samples were used for training in each category, and the remaining individual samples were used for testing. The design destination uses limited training samples to test the generalization performance of different classifiers.
The performance of the different classifiers was compared in terms of class specific precision, overall precision (OA), average precision (AA), Kappa coefficient (κ), and the differences between the classifiers were evaluated using Kappa and McNemar z-value statistical tests.
Therefore, by selecting validation samples based on visual interpretation of GlobeLand30 covering the same area, to reduce the impact caused by time and artifacts during sample selection, all interpreted samples were screened using Vertex Component Analysis (VCA), step D design application; vertex Component Analysis (VCA) is unsupervised and is based on the geometry of the convex set, which exploits the fact that the end-members occupy the vertices of a simplex. VCA is superior to PPI, superior to or similar to N-FINDR. And VCA has the lowest computational complexity of the three algorithms. The savings in computational complexity are between one and two orders of magnitude. Thus, here VCA is used to generate cleaner samples that can represent a particular type of land cover well.
And in practical applications, the comparison of the accuracy obtained by each classifier, for example, is shown in table 1 below.
TABLE 1
k-NN: k nearest neighbor classifiers (Cover and Hart, 1967).
SVM: support vector machines (cortex and Vapnik, 1995).
LORSAL: via variable splitting and augmented lagrange polynomial regression (Li et al, 2011).
SRC: sparse representative classification (Wright et al, 2009).
Bagging: a learning-based ensemble classifier formed by bootstrap replication of training data (Breiman, 1996).
RoF: the forest is rotated, and set-learning based classifiers generated from the basic classifiers are extracted based on features (Rodriguez et al, 2006). And RS: random subspace, an ensemble learning approach, which attempts to reduce the correlation between estimators in a set by training random samples of features rather than the entire set of features. (Ho, 1998).
RF: random forest, a collection of tree predictors, such that each tree depends on randomly selected features (Breiman, 2001). The cutoff value for z-score at 0.95 level was 1.96. All tests showed 95% performance with the exception of Bagging/RF.
And E, applying a target classifier to classify each pixel region in the historical vegetation sample image sequence corresponding to the target region to obtain the target object type corresponding to each pixel region in the historical vegetation sample image sequence corresponding to the target region, and then entering the step F.
And F, aiming at the historical vegetation sample image sequence corresponding to the target area, applying STL decomposition to remove peaks, extracting seasonal parameters in the historical vegetation sample image sequence, applying a Savitzky-Golay filter to perform fitting filtering aiming at the time of the historical vegetation sample image sequence, updating the historical vegetation sample image sequence, and then entering the step G.
In practical applications, the step F includes steps F1 to F2.
Step F1, aiming at a historical vegetation sample image sequence corresponding to a target area, performing peak removal by using STL decomposition, and extracting seasonal parameters in the historical vegetation sample image sequence, namely aiming at the historical vegetation sample image sequence, decomposing the historical vegetation sample image sequence into seasons, trends and remainders as follows;
Yt＝St+Tt+et (1)
where t is 1, …, n, n represents the total number of time nodes corresponding to the image sequence of the historical vegetation sample, and Y represents the total number of time nodes corresponding to the image sequence of the historical vegetation sampletIs the data of the corresponding time node t in the image sequence of the historical vegetation sample StRepresenting seasonal components, T, corresponding to the decomposition of the corresponding time node T in the image sequence of the historical vegetation samplestRepresenting trend components, e, corresponding to the decomposition of the corresponding time node t in the image sequence of the historical vegetation sampletThe residue component corresponding to the decomposition of the corresponding time node t in the historical vegetation sample image sequence is shown.
Wherein, m seasonal breakpoints in the image sequence are preset according to the historical vegetation samples
And define
Based on j being 1, …, m, StThe corresponding harmonic model with the K harmonic term is represented as follows:
wherein m represents the number of seasonal breakpoints in the preset historical vegetation sample image sequence, aj,kRepresenting the amplitude, delta, of the image sequence of the historical vegetation sample corresponding to the jth seasonal breakpoint and the kth harmonic analysisj,kRepresenting the phase of the image sequence of the historical vegetation sample corresponding to the jth seasonal breakpoint and the kth harmonic wave analysis, and f representing each calendar of each seasonal breakpoint corresponding to the image sequence of the historical vegetation sampleFrequency of image segments of the vegetation samples.
According to TtAnd the slope beta of each historical vegetation sample image segment of each seasonal breakpoint corresponding to the historical vegetation sample image sequenceiIntercept αiObtaining TtIs represented as follows:
Tt＝αi+βit (3)
where i is 1, …, m +1, i.e. m +1 image segments of the historical vegetation sample, βiRepresenting the slope, α, of the image segment of the ith historical vegetation sampleiRepresenting the intercept of the image segment representing the ith historical vegetation sample.
Step F2. is based on the decomposition of the historical vegetation sample image sequence with respect to season, trend, and residue, applying Savitzky-Golay filter, in combination with the sliding of the window in the historical vegetation sample image sequence, as follows:
the average value p of the linear combination of the window neighborhood values is used for each data y in the windowqThe fitting filtering to the image sequence of the historical vegetation sample is realized, wherein Q is 1, …, Q represents the number of data in the window, yqRepresenting the q-th data within the window, cgIs represented by cg＝1/(2n+1)。
The critical points, lines and regions which can be determined by the phenological model are referred to as phenological indicators of seasonal tracks. For example, the start of season (SOS) is one of the most common indicators, meaning the maximum half time during which vegetation is growing. In addition, certain other indicators, such as end of season (EOS), length of season (LOS), base of season level (BOS), mid season (TOMS), peak season (POS), amplitude of season (AOS), growth Rate (ROG), aging Rate (ROS), Gross Spring Greenness (GSG) and Net Spring Greenness (NSG) were also included in our studies based on time series of tiseat fits, such as the definitions of the different weather indicators shown in table 2 below.
TABLE 2
And G, marking the phenological information of the historical vegetation sample image sequence by using a TIMEAT time sequence analysis method based on the target object type corresponding to each pixel region in the historical vegetation sample image sequence corresponding to the target region, and obtaining the distribution of each phenological mark on the target region space and the evolution rule of each phenological mark along with seasonal parameters on a time sequence.
Based on the mode, the analysis is carried out on the Yangtze river delta area, the forest area shows clear and early SOS, and the cultivated land shows obvious and late SOS. Wetlands exhibited earlier SOS (i.e., 0-38). Most regions have similar EOSs, with very little difference between the north and south regions of the Yangtze delta. The LOS in the north is less than the LOS in the south. Different types of mulch show significantly different BOS, with forests having higher BOS than cultivated land. TOMS values are different in different regions, TOOS is smaller in coastal regions of east coast, and SOS and EOS standards are met. Similar POS were present in most of the study area except for the water and artificial surfaces, while lower AOS were present in the southern area. South shows higher ROG and ROS, indicating vegetation vigor in forests in the area. GSG and NSG reflect regional spring greenness and are directly related to biomass. The corresponding results indicate that the north of the Yangtze river delta has greater productivity because the cultivated land is distributed in the north according to the classification chart.
The phenological extraction and ground surface coverage classification method based on MODIS long-time sequence data is designed by the technical scheme, a historical vegetation sample image sequence is obtained, then the historical vegetation sample image sequence corresponding to a target area is subjected to batch processing by adopting an MRT tool, the historical vegetation sample image sequence corresponding to the target area is updated, then GlobalLand30 is applied to obtain images of all target object types in the historical vegetation sample image sequence corresponding to the target area and serve as all samples, a vertex component analysis method is applied to screen all samples, a classifier is trained to obtain a target classifier, then the target classifier is applied to classify to obtain the target object types corresponding to all pixel areas in the historical vegetation sample image sequence corresponding to the target area respectively, extraction of seasonal parameters in the historical vegetation sample image sequence is combined, and performing fitting filtering on the time of the historical vegetation sample image sequence, updating the historical vegetation sample image sequence, and finally obtaining the distribution of each phenological mark on a target area space and the evolution rule of each phenological mark along with seasonal parameters on the time sequence.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (5)
1. A phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data is characterized by comprising the following steps:
a, obtaining an MODIS global vegetation index image sequence which corresponds to a target area and has a preset historical time range and contains NDVI and EVI, taking the obtained MODIS global vegetation index image sequence as a historical vegetation sample image sequence corresponding to the target area, and then entering the step B;
b, adopting an MRT tool to perform batch processing on the historical vegetation sample image sequence corresponding to the target area, converting the Sinusoid projection in the historical vegetation sample image sequence into a preset target space coordinate system, updating the historical vegetation sample image sequence corresponding to the target area, and entering the step C;
step C, obtaining images of each target object type in the historical vegetation sample image sequence corresponding to the target area by using GlobalLand30 as each sample, and then entering step D;
d, screening all samples by using a vertex component analysis method to obtain each training sample, training the training samples by using the training samples as input and the target object types corresponding to the training samples as output, training each preset classifier to be trained to obtain a trained classifier corresponding to the highest accuracy as a target classifier, and then entering the step E;
e, classifying each pixel region in the historical vegetation sample image sequence corresponding to the target region by applying a target classifier to obtain target object types respectively corresponding to each pixel region in the historical vegetation sample image sequence corresponding to the target region, and then entering the step F;
f, aiming at the historical vegetation sample image sequence corresponding to the target area, performing peak elimination by using STL decomposition, extracting seasonal parameters in the historical vegetation sample image sequence, performing fitting filtering on the time of the historical vegetation sample image sequence by using a Savitzky-Golay filter, updating the historical vegetation sample image sequence, and then entering the step G;
and G, marking the phenological information of the historical vegetation sample image sequence by using a TIMEAT time sequence analysis method based on the target object type corresponding to each pixel region in the historical vegetation sample image sequence corresponding to the target region, and obtaining the distribution of each phenological mark on the target region space and the evolution rule of each phenological mark along with seasonal parameters on a time sequence.
2. The method for phenological extraction and earth surface coverage classification based on MODIS long-time sequence data according to claim 1, wherein: in the step B, the MRT tool is used to batch process the image sequence of the historical vegetation sample corresponding to the target area, the sinussoid projection in the image sequence of the historical vegetation sample is converted into WGS _1984_ UTM _ Zone _50N, and the image sequence of the historical vegetation sample corresponding to the target area is updated.
3. The method for phenological extraction and earth surface coverage classification based on MODIS long-time sequence data according to claim 1, wherein: and in the step C, obtaining images of target object types of cultivated land, forest, grassland, wetland, water body and artificial surface in the historical vegetation sample image sequence corresponding to the target area by using GlobalLand30 as each sample.
4. The method for phenological extraction and earth surface coverage classification based on MODIS long-time sequence data according to claim 1, wherein: and D, presetting each classifier to be trained to comprise a rotary forest classifier, a Bagging classifier and a random forest classifier.
5. The method for phenological extraction and earth surface coverage classification based on MODIS long-time sequence data according to claim 1, wherein said step F includes steps F1 to F2 as follows:
step F1, aiming at a historical vegetation sample image sequence corresponding to a target area, performing peak removal by using STL decomposition, and extracting seasonal parameters in the historical vegetation sample image sequence, namely aiming at the historical vegetation sample image sequence, decomposing the historical vegetation sample image sequence into seasons, trends and remainders as follows;
Yt＝St+Tt+et (1)
in the formula, t is 1, n, n represents the number of all time nodes corresponding to the image sequence of the historical vegetation sample, and Y represents the number of all time nodes corresponding to the image sequence of the historical vegetation sampletIs the data of the corresponding time node t in the image sequence of the historical vegetation sample StRepresenting seasonal components, T, corresponding to the decomposition of the corresponding time node T in the image sequence of the historical vegetation samplestRepresenting trend components, e, corresponding to the decomposition of the corresponding time node t in the image sequence of the historical vegetation sampletRepresenting a remainder component corresponding to the decomposition of the corresponding time node t in the historical vegetation sample image sequence;
wherein, m seasonal breakpoints in the image sequence are preset according to the historical vegetation samples
And define
Then based on j ═ 1.., m, StThe corresponding harmonic model with the K harmonic term is represented as follows:
wherein m represents the number of seasonal breakpoints in the preset historical vegetation sample image sequence, aj，kRepresenting the amplitude, delta, of the image sequence of the historical vegetation sample corresponding to the jth seasonal breakpoint and the kth harmonic analysisj，kRepresenting the phase of the image sequence of the historical vegetation sample corresponding to the jth seasonal breakpoint and the phase under the kth harmonic analysis, and f representing the frequency of each image segment of the historical vegetation sample corresponding to each seasonal breakpoint of the image sequence of the historical vegetation sample;
according to TtAnd the slope beta of each historical vegetation sample image segment of each seasonal breakpoint corresponding to the historical vegetation sample image sequenceiIntercept αiObtaining TtIs represented as follows:
Tt＝αi+βit (3)
where i is 1., m +1, i.e., m +1 image segments of the historical vegetation sample, βiRepresenting the slope, α, of the image segment of the ith historical vegetation sampleiRepresenting an intercept representing an image segment of the ith historical vegetation sample;
step F2. is based on the decomposition of the historical vegetation sample image sequence with respect to season, trend, and residue, applying Savitzky-Golay filter, in combination with the sliding of the window in the historical vegetation sample image sequence, as follows:
the average value p of the linear combination of the window neighborhood values is used for each data y in the windowqImplementing fitting filtering on the image sequence of the historical vegetation sample, wherein Q is 1qWithin a presentation windowData of q number, cgIs represented by cg＝1/(2n+1)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111233559.5A CN114120027A (en) | 2021-10-22 | 2021-10-22 | Phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111233559.5A CN114120027A (en) | 2021-10-22 | 2021-10-22 | Phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114120027A true CN114120027A (en) | 2022-03-01 |
Family
ID=80376747
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111233559.5A Pending CN114120027A (en) | 2021-10-22 | 2021-10-22 | Phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114120027A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117910539A (en) * | 2024-03-19 | 2024-04-19 | 电子科技大学 | Household characteristic recognition method based on heterogeneous semi-supervised federal learning |
-
2021
- 2021-10-22 CN CN202111233559.5A patent/CN114120027A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117910539A (en) * | 2024-03-19 | 2024-04-19 | 电子科技大学 | Household characteristic recognition method based on heterogeneous semi-supervised federal learning |
| CN117910539B (en) * | 2024-03-19 | 2024-05-31 | 电子科技大学 | Household characteristic recognition method based on heterogeneous semi-supervised federal learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Blickensdörfer et al. | Mapping of crop types and crop sequences with combined time series of Sentinel-1, Sentinel-2 and Landsat 8 data for Germany | |
| Alonzo et al. | Urban tree species mapping using hyperspectral and lidar data fusion | |
| Laurin et al. | Discrimination of tropical forest types, dominant species, and mapping of functional guilds by hyperspectral and simulated multispectral Sentinel-2 data | |
| Phinn et al. | Monitoring the composition of urban environments based on the vegetation-impervious surface-soil (VIS) model by subpixel analysis techniques | |
| Herold et al. | Some challenges in global land cover mapping: An assessment of agreement and accuracy in existing 1 km datasets | |
| Borak | Feature selection and land cover classification of a MODIS-like data set for a semiarid environment | |
| CN103632363B (en) | Object level high-resolution remote sensing image change detecting method based on Multiscale Fusion | |
| CN109711446A (en) | A kind of terrain classification method and device based on multispectral image and SAR image | |
| Chen | Mapping mangrove in Dongzhaigang, China using Sentinel-2 imagery | |
| Guyet et al. | Long term analysis of time series of satellite images | |
| CN104794424A (en) | Novel middle-low resolution remote sensing data combined cultivated land identification method | |
| Zhang et al. | Spatial domain bridge transfer: An automated paddy rice mapping method with no training data required and decreased image inputs for the large cloudy area | |
| Zhang et al. | Mapping freshwater marsh species in the wetlands of Lake Okeechobee using very high-resolution aerial photography and lidar data | |
| Wang et al. | Spatio-temporal analysis of phenology in Yangtze River Delta based on MODIS NDVI time series from 2001 to 2015 | |
| Wang et al. | Decadal trend in agricultural abandonment and woodland expansion in an agro-pastoral transition band in Northern China | |
| CN117475313A (en) | Large-scale winter wheat early-stage identification method based on automatic extraction of training samples | |
| Billah et al. | Random forest classifications for landuse mapping to assess rapid flood damage using Sentinel-1 and Sentinel-2 data | |
| CN113963261A (en) | Method and system for extracting full convolution neural network cultivated land based on multi-scale fusion | |
| CN114120027A (en) | Phenological extraction and earth surface coverage classification method based on MODIS long-time sequence data | |
| Zhu et al. | Identification of soybean based on Sentinel-1/2 SAR and MSI imagery under a complex planting structure | |
| Zhao et al. | Decision surface optimization in mapping exotic mangrove species (Sonneratia apetala) across latitudinal coastal areas of China | |
| Du et al. | An ensemble learning approach for land use/land cover classification of arid regions for climate simulation: A case study of xinjiang, northwest China | |
| Tokar et al. | Using the Random Forest classification for land cover interpretation of Landsat images in the Prykarpattya region of Ukraine | |
| Paris et al. | A scalable high-performance unsupervised system for producing large-scale hr land cover maps: The italian country case study | |
| Soh et al. | Automated near-real-time mapping and monitoring of rice growth extent and stages in Selangor Malaysia |
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 |