CN117292154B - Automatic production method of long-time-sequence ground object samples based on dense time-sequence remote sensing images - Google Patents
Automatic production method of long-time-sequence ground object samples based on dense time-sequence remote sensing images Download PDFInfo
- Publication number
- CN117292154B CN117292154B CN202311579977.9A CN202311579977A CN117292154B CN 117292154 B CN117292154 B CN 117292154B CN 202311579977 A CN202311579977 A CN 202311579977A CN 117292154 B CN117292154 B CN 117292154B
- Authority
- CN
- China
- Prior art keywords
- time
- remote sensing
- time period
- sequence
- ground object
- 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
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 26
- 238000001228 spectrum Methods 0.000 claims abstract description 108
- 230000003595 spectral effect Effects 0.000 claims abstract description 93
- 238000000034 method Methods 0.000 claims description 44
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 16
- 238000005520 cutting process Methods 0.000 claims description 4
- 238000003809 water extraction Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 description 18
- 230000000007 visual effect Effects 0.000 description 9
- 230000008569 process Effects 0.000 description 7
- 238000012795 verification Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/62—Extraction of image or video features relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
-
- 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
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- 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
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention provides a long-time-sequence ground object sample automatic production method based on dense time-sequence remote sensing images, which belongs to the technical field of remote sensing algorithms and comprises the following steps: determining a first association degree between a spectral feature time sequence of a first remote sensing image of a sample point in a first time period and a spectral feature time sequence of a second remote sensing image of the sample point in a second time period; when the first association degree is larger than a preset threshold value, taking the ground object type of the sample point in the first time period as the ground object type of the sample point in the second time period; otherwise, determining the ground object type of the sample point in the second time period according to the second association degree between the spectrum characteristic time sequence of the second remote sensing image of the sample point in the second time period and the preset spectrum characteristic time sequence of the ground object type. According to the invention, the ground object type of the sample point in the first time period and the spectrum characteristic time sequence are taken as references to identify the ground object type of the sample point in the second time period, so that the automatic production of the long-time sequence ground object sample is realized, and manual intervention is not required.
Description
Technical Field
The invention relates to the technical field of remote sensing algorithms, in particular to an automatic production method of a long-time-sequence ground object sample based on a dense time-sequence remote sensing image.
Background
Sample data production is one of important basic works for developing land cover remote sensing drawing, and is mainly used for training supervision classifiers, mining classification knowledge and evaluating drawing precision. The number, accuracy and representativeness of the sample data are important indicators for measuring the quality of the sample. For the land cover remote sensing drawing research, the method requires sufficient number of land feature samples, high data precision and reasonable spatial distribution.
The production of the ground object sample refers to determining the ground object category attribute of a certain position point of a target area through a certain technical means.
At an early stage, land feature sample production is mostly carried out using field measurement and visual interpretation. The field measurement method is to determine a pure sample in a single land use type range with larger plaque; and then using a global positioning system (Global Position System, GPS), a handheld PAD, a total station and other measuring equipment to carry out coordinate measurement on the sample points and recording the category attributes of the sample points. The visual interpretation method is to superimpose each sample point on the remote sensing image by using a high spatial resolution satellite image, an aerial photograph, an unmanned aerial vehicle photograph and the like, and visually interpret the category attribute of the sample point by an interpreter. The ground object samples produced by the two methods have high precision and are mainly used for early ground object sample data production.
However, when field measurement or visual interpretation is used to produce the ground object samples, a lot of manpower is required to complete field mapping or identify a lot of visual interpretation work, and meanwhile, a lot of time and economic cost are required, so that it is difficult to generate ground object samples with large area and long time sequence.
Disclosure of Invention
The invention provides a long-time-sequence ground object sample automatic production method based on a dense time-sequence remote sensing image, which is used for solving the defect of high production cost of early-stage ground object samples and realizing an efficient and automatic ground object sample production method.
The invention provides a long-time-sequence ground object sample automatic production method based on dense time-sequence remote sensing images, which comprises the following steps:
determining a first association degree between a spectral feature time sequence of a first remote sensing image of a sample point in a target area in a first time period and a spectral feature time sequence of a second remote sensing image of the sample point in a second time period;
taking the ground object type of the sample point in a first time period as the ground object type of the sample point in a second time period under the condition that the first association degree is larger than a preset threshold value;
determining a second degree of correlation between the spectral feature time sequence of the second remote sensing image of the sample point in a second time period and a preset spectral feature time sequence of at least one ground object type under the condition that the first degree of correlation is smaller than or equal to the preset threshold;
and taking the ground object type corresponding to the second association degree with the maximum degree as the ground object type of the sample point in the second time period.
According to the method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images, the step of determining the first degree of correlation between the spectral feature time sequence of a first remote sensing image of a sample point in a target area and the spectral feature time sequence of a second remote sensing image of the sample point in a second time period comprises the following steps:
determining a third association degree between each spectrum characteristic time sequence of the sample point in the first remote sensing image of the first time period and the spectrum characteristic time sequence of the corresponding kind of the sample point in the second remote sensing image of the second time period;
and determining the first association degree according to the third association degree.
According to the method for automatically producing the long-time-sequence ground object samples based on the dense time-sequence remote sensing images, the step of determining the second degree of correlation between the spectrum characteristic time sequence of the second remote sensing image of the sample point in the second time period and the preset spectrum characteristic time sequence corresponding to at least one ground object type comprises the following steps:
determining a fourth association degree between the spectrum characteristic time sequence of the sample point in the second remote sensing image of the second time period and the preset spectrum characteristic time sequence of the corresponding type of each ground object type;
and determining a second degree of correlation between the spectral feature time sequence of the second remote sensing image of the sample point in the second time period and a preset spectral feature time sequence corresponding to at least one ground object type according to the fourth degree of correlation.
According to the method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images provided by the invention, before the step of determining the second degree of correlation between the spectrum characteristic time sequence of the second remote sensing image in the second time period and the preset spectrum characteristic time sequence corresponding to at least one ground object type, the method further comprises the following steps:
determining a spectrum characteristic average value of a spectrum characteristic time sequence of a first remote sensing image of the sample point belonging to the same ground object type in a first time period at each moment;
integrating the spectrum characteristic average value at each moment belonging to the same feature type sample point into a spectrum characteristic time sequence of the corresponding feature type, and taking the spectrum characteristic time sequence as a preset spectrum characteristic time sequence of the feature type;
and integrating the preset spectrum characteristic time sequence of each ground object type to obtain the preset spectrum characteristic time sequence corresponding to the at least one ground object type.
According to the method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images, before the step of determining the first degree of correlation between the spectral feature time sequence of the first remote sensing image of the sample point in the target area and the spectral feature time sequence of the second remote sensing image of the sample point in the second time period, the method further comprises the steps of:
after removing pixels which are shielded by cloud layers or cloud shadows in the remote sensing images at each moment in the first time period and the second time period, embedding and cutting the remote sensing images at each moment in the first time period and the second time period to obtain the first remote sensing image and the second remote sensing image, wherein the boundaries of the first remote sensing image and the second remote sensing image are consistent with the boundaries of the target areas to which the sample points belong.
According to the automatic production method of the long-time-sequence ground object samples based on the dense time sequence remote sensing images, the spectral characteristics corresponding to the spectral characteristic time sequence comprise one or more of green wave bands, red wave bands, near infrared wave bands, short wave infrared 1, short wave infrared 2, normalized vegetation indexes, normalized water body indexes, improved normalized water body indexes and automatic water body extraction indexes of the remote sensing images.
According to the automatic production method of the long-time-sequence ground object sample based on the dense time-sequence remote sensing image, provided by the invention, the ground object type of a sample point of a target area in a target time period is identified by taking the known remote sensing image time sequence of the target area in a certain time period as a reference and combining a predetermined spectral characteristic time sequence corresponding to each ground object type on the basis of remote sensing big data and an image processing algorithm and on the basis of the assumption that the same ground object type has the same or similar spectral characteristics, so that the automatic production of the long-time-sequence ground object sample is realized without manual intervention.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images;
FIG. 2 is a second flow chart of the method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images;
FIG. 3 is a third flow chart of the method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The following describes a method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images with reference to fig. 1 to 3, wherein the method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images comprises the following steps:
step 101, determining a first association degree between a spectral feature time sequence of a first remote sensing image of a sample point in a target area in a first time period and a spectral feature time sequence of a second remote sensing image of the sample point in a second time period;
the target area is an area corresponding to the ground object sample, and the sample point is positioned in the target area.
The first time period is one day, one week and the like, the first time period comprises a plurality of moments, and the time length of each moment can be determined according to the time length of the first time period.
Alternatively, the time interval per time within the first period may be the same or different.
In the invention, the interval between each time is 16 days, and the remote sensing image of the target sample point in the first time period comprises Landsat satellite remote sensing images of the target region in the first time period every 16 days.
On this basis, optionally, in the case that the duration of the first time period is one year, the first time period may be any natural year, and when the natural year is not leap year, the first time period may include 365 times; when the natural year is leap year, the first time period may include 366 moments.
Alternatively, in the case where the first period of time is one year, the first period of time may also be any specified date of any year to the corresponding date of the next year.
And after determining the start and stop date of the first time period and the time interval of each moment of the first time period according to the requirements, correspondingly acquiring the first remote sensing image of the target area in each moment of the first time period.
Optionally, the first remote sensing image is a Landsat image (a remote sensing image acquired by a Landsat satellite system).
Specifically, aiming at a target area, using a Google Earth Engine remote sensing big data cloud platform, searching all Landsat ground surface reflectivity images of the target area in a first time period, including images of Landsat 4/5 TM, landsat 7 ETM+, landsat 8/9 OLI (all being different satellites of a Landsat satellite system), and sequencing the remote sensing images in the searched all first time periods according to the sequence of time to obtain a remote sensing image time sequence of the target area in the first time period.
The second time period corresponds to the first time period, and the time division in the second time period also corresponds to the time division in the first time period, so that the remote sensing image time sequence of the target area in the second time period is obtained.
Wherein the first time period represents a reference time period in which the type of the ground feature of the sample point in the target area is known; the second time period represents a target time period, and the ground object type of the sample point in the target area in the target time period is to be identified.
Alternatively, the first time period may precede the second time period, or may follow the second time period.
The first remote sensing image and the second remote sensing image at each moment in the first time period and the second time period have corresponding spectral characteristics, and the spectral characteristics of the first remote sensing image and the second remote sensing image at each moment form a spectral characteristic time sequence of the first remote sensing image and a spectral characteristic time sequence of the second remote sensing image.
And determining a first association degree between the spectral feature time sequence of the first remote sensing image of the sample point of the target area and the spectral feature time sequence of the sample point of the second remote sensing image.
Wherein the first degree of correlation represents the degree of similarity of the spectral features of the sample points in the first time period and the second time period. Therefore, the similarity degree of the types of the features of the corresponding sample points can be determined by comparing the similarity degree of the spectral features.
And if the spectrum characteristic time sequence of the first remote sensing image of the sample point in the first time period is similar to the spectrum characteristic time sequence of the second remote sensing image in the second time period, the type of the ground feature of the sample point in the first time period is considered to be the same as the type of the ground feature of the sample point in the second time period.
On the basis, the type of the ground object in the second time period can be directly determined through the type of the ground object in the first time period at the sample point.
Among these types of land features include, but are not limited to, bodies of water, herbaceous swamps, woody swamps, inland beach lands, coastal beach lands, woodlands, grasslands, construction lands, paddy fields, dry lands, bare lands.
Step 102, taking the ground object type of the sample point in a first time period as the ground object type of the sample point in a second time period under the condition that the first association degree is larger than a preset threshold value;
the preset threshold is empirically determined.
If the first association degree is greater than the preset threshold value, the spectrum characteristic time sequence of the first remote sensing image of the sample point in the first time period is similar to the spectrum characteristic time sequence of the second remote sensing image in the second time period, namely the type of the ground object at the pixel position in the remote sensing image corresponding to the sample point is unchanged, and the sample point is regarded as an annual stable sample point in the application.
Therefore, the ground object type of the annual stable sample point in the first time period is used as the ground object type of the annual stable sample point in the second time period, and further the ground object type attribute identification of the annual stable sample point in the second time period is realized.
Step 103, determining a second degree of correlation between the spectral feature time sequence of the second remote sensing image of the sample point in a second time period and a preset spectral feature time sequence of at least one ground object type under the condition that the first degree of correlation is smaller than or equal to the preset threshold;
if the first association degree is smaller than the preset threshold value, it means that compared with the first time period, the spectrum characteristics of the sample point in the second time period are changed, that is, the feature type corresponding to the sample point is changed along with the change of time, and the sample point is regarded as an annual change sample point in the application.
For the annual change sample points, the corresponding ground object categories need to be further identified.
Specifically, comparing the time sequence of the spectrum characteristic of the internationally-changed sample point in the second remote sensing image with the time sequence of the preset spectrum characteristic corresponding to at least one ground feature type, and determining a second association degree between the time sequence of the spectrum characteristic of the internationally-changed sample point and the preset spectrum characteristic value corresponding to at least one ground feature type.
The preset spectrum characteristic time sequence corresponding to the at least one ground object type is a predetermined standard spectrum characteristic time sequence of the at least one ground object type.
Therefore, the second degree of association indicates the degree of similarity between the annual change sample point and each of the feature types, and the larger the second degree of association, the closer the feature type corresponding to the annual change sample point is to the feature type corresponding to the second degree of association.
The method comprises the steps of determining a spectrum characteristic time sequence average value of a first remote sensing image with a plurality of sample points of the same ground object type in a first time period, wherein the preset spectrum characteristic time sequence corresponding to at least one ground object type is determined by the spectrum characteristic time sequence average value of the first remote sensing image with a plurality of sample points of the same ground object type in the first time period.
And 104, taking the ground object type corresponding to the second association degree with the maximum degree as the ground object type of the sample point in the second time period.
And for each time-series change sample point, arranging the second association degree between the time-series change sample points and the preset spectrum time sequences corresponding to all the time-series change sample points, and taking the time-series change sample point with the highest second association degree as the time-series change sample point to finish the time-series change sample point.
On the basis, the ground object types of all sample points of the target area in the second time period are determined, and then the production of the long-time-sequence ground object samples of the target area in the second time period is completed.
According to the invention, the type of the ground object of the sample point of the target area in the target time period is identified by taking the known remote sensing image time sequence of the target area in a certain time period as a reference and combining the predetermined spectrum characteristic time sequence corresponding to each ground object type based on the assumption that the same ground object type has the same or similar spectrum characteristics by taking the remote sensing big data and the image processing algorithm as the basis, so that the automatic production of the long-time sequence ground object sample is realized without manual intervention.
In the automatic production of the long-time-sequence ground object sample based on the dense time sequence remote sensing image, the spectral characteristics corresponding to the spectral characteristic time sequence comprise one or more of a green wave band, a red wave band, a near infrared wave band, a short wave infrared 1, a short wave infrared 2, a normalized vegetation index, a normalized water index, an improved normalized water index and an automatic water extraction index of the remote sensing image.
The spectral features corresponding to the spectral feature time sequence are the spectral features in the remote sensing image at each moment.
In the remote sensing image, a plurality of spectral features are provided, and the value corresponding to the spectral features is a spectral feature value.
Spectral features first include the Green band (Green), red band (Red), near infrared band (NIR), short wave infrared 1 (SWIR 1) and short wave infrared 2 (SWIR 2) of the target region Landsat image.
The spectral characteristics also comprise normalized vegetation index (NDVI), normalized water index (NDWI), improved normalized water index (MDNWI) and automatic water extraction index (AWI), and the four indexes are calculated by corresponding values of each wave band of the Landsat image, and the specific calculation formula is as follows:
（1）
（2）
（3）
（4）
according to the invention, on the basis of the green wave band, the red wave band, the near infrared wave band, the short wave infrared 1 and the short wave infrared 2 of the remote sensing image, the normalized vegetation index, the normalized water index, the improved normalized water index and the automatic water extraction index are combined to serve as the spectral characteristics corresponding to the ground object types, so that the recognition of vegetation information and water information of the remote sensing image can be enhanced, and the accuracy of the recognition of the ground object remote sensing image can be improved.
In the automatic production of the long-time-sequence ground object sample based on the dense time-sequence remote sensing image, the invention determines a first association degree between a spectrum characteristic time sequence of a first remote sensing image of a sample point in a target area and a spectrum characteristic time sequence of a second remote sensing image of the sample point in a second time period, and comprises the following steps:
determining a third association degree between each spectrum characteristic time sequence of the sample point in the first remote sensing image of the first time period and the spectrum characteristic time sequence of the corresponding kind of the sample point in the second remote sensing image of the second time period;
because there are a plurality of spectral features of the remote sensing images, a third degree of correlation between each spectral feature time sequence of the target sample point in the first remote sensing image in the first time period and a corresponding kind of spectral feature time sequence in the second remote sensing image corresponding to the first time period needs to be determined.
Specifically, firstly, obtaining a spectrum characteristic value of each spectrum characteristic of a first remote sensing image of a sample point in the target area at each moment in a first time period, and generating a spectrum characteristic time sequence according to each spectrum characteristic value corresponding to the sample point in time sequence.
In a possible embodiment, a coordinate system is established for each spectral feature of each sample point, wherein the abscissa represents the point in time, in this case 1 to 365, and the ordinate represents the spectral feature value of that spectral feature of that sample point.
On the basis, a spectral feature time sequence of the second remote sensing image of each sample point in the second time period is constructed in the same way. And calculating the association degree between the two corresponding time sequences by using a gray association model.
In one possible embodiment, as shown in fig. 2, the time series image of the reference year in fig. 2 is a time series of spectral features corresponding to the first remote sensing image, and the time series image of the target year is a time series of spectral features corresponding to the second remote sensing image.
For any sample point of the target area, the time sequence of the spectral characteristics of the target area in the first time period and the second time period is respectively set asAnd->The expression is as follows:
（5）
（6）
the sample point is provided with pixels corresponding to the first remote sensing image and the second remote sensing image, and the pixels identify the ground object type corresponding to the obtained spectral characteristics, namely the ground object type of the sample point in the target area.
In the method, in the process of the invention,codes representing spectral features, corresponding to Blue, green, red, NIR, SWIR, SWRI2, NDVI, NDWI, MNDWI, AWEI; />Representing the +.f. of the first remote sensing image>Time series of individual spectral features,/->Second remote sensing image +.>Time series of individual spectral features；kRepresenting time node [ ]Day of Year，Doy) I.e. every moment in time within the first time period and the second time period;nindicating the total number of time sequences.
Calculating the absolute difference, the minimum absolute difference value and the maximum absolute difference value of each spectrum characteristic time sequence of each sample point in the second time period and the first time:
（7）
（8）
（9）
in the method, in the process of the invention,first and second time periods ∈>Characterised by the spectrumkAbsolute difference of time,/->、/>Respectively represent the minimum value and the maximum value of the time series corresponding to the absolute difference value.
Further, calculating an association coefficient corresponding to a certain spectral feature of each sample point at a certain corresponding moment:
（10）
in the method, in the process of the invention,first and second time periods ∈>Characterised by the spectrumkThe correlation coefficient of the time of day,θfor the resolution factor, the value is usually 0.5.
And then according to the association coefficient, calculating a third association degree between each spectrum characteristic time sequence of each sample point in the first remote sensing image in the first time period and each spectrum characteristic time sequence of the corresponding type of the sample point in the second remote sensing image in the second time period:
（11）
in the method, in the process of the invention,first ∈h represented as first period and second period>And determining the association degree between each spectrum characteristic time sequence of the first time period and the second time period as a third association degree.
And determining the first association degree according to the third association degree.
And after carrying out weighted summation on the third association degree corresponding to each spectrum characteristic time sequence of each sample point, obtaining the total association degree corresponding to the sample point, namely the first association degree:
（12）
in the method, in the process of the invention,Rfor the first degree of association,mas a function of the total number of spectral features,is->Weighting coefficients for each spectral feature, wherein the weighting coefficients for each spectral feature are empirically determined。
Through the calculation, the first association degree corresponding to the sample point can be determined based on the multiple spectral feature time sequences of the first remote sensing image of the sample point in the first time period and the spectral feature time sequences of the corresponding types of the second remote sensing image of the sample point in the second time period, so that the sample point is identified as an annual stable sample point or an annual change sample point, and the ground object type of the sample point is further determined.
The gray correlation model can describe the continuous change process of things by using fewer original data sequences under the condition of partial data missing. The degree of matching between sample points can be well characterized by calculating the degree of correlation based on the gray correlation model. Under the condition that the time series data of the sample points are sparse, the first association degree calculated by the method can describe the similarity between the sample points more stably, and has better adaptability in complex scenes.
According to the method, the first association degree between the spectral feature time sequence of the first remote sensing image and the spectral feature time sequence of the second remote sensing image in the first time period of the sample points in the target area is calculated based on the gray association model, automatic matching identification of the sample points between the second time period and the first time period is achieved, the problem that the matching precision of the sample points of single-phase spectral features is low is solved, and even if partial point location data is incomplete, the change trend is nonlinear and the like in the spectral feature time sequence corresponding to the first remote sensing image, the first association degree of each sample point can be determined through the gray association model.
The invention discloses a method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images, which comprises the following steps of determining a second association degree between a spectrum characteristic time sequence of a second remote sensing image of a sample point in a second time period and a preset spectrum characteristic time sequence corresponding to at least one ground object type, wherein the second association degree comprises the following steps:
determining a fourth degree of association between each spectrum characteristic time sequence of the sample point in the second remote sensing image of the second time period and a preset spectrum characteristic time sequence of a corresponding type of each ground object type;
and determining a second degree of correlation between the spectral feature time sequence of the second remote sensing image of the sample point in the second time period and a preset spectral feature time sequence corresponding to at least one ground object type according to the fourth degree of correlation.
The method for determining the fourth degree of correlation between each spectral feature time series of each sample point in the second time period and each preset spectral feature value corresponding to each ground feature type is approximately the same as the method for determining the third degree of correlation between each spectral feature time series of each sample point in the first time period and the corresponding spectral feature time series of the corresponding sample point in the second time period.
Specifically, each preset spectrum characteristic time sequence corresponding to each ground object type is determined, and the time sequence is represented in the same manner as the spectrum characteristic time sequences corresponding to the first time period and the second time period, so that the description is omitted. The preset spectrum characteristic time sequence is a standard spectrum characteristic time sequence of the ground object type.
On the basis, a fourth association degree is calculated by using the formulas (5) to (11), a second association degree is calculated by using the formula (12) according to the fourth association degree, and the spectral feature time sequence of the sample points in the formula in the first time period is replaced by the preset spectral feature time sequence of at least one ground object type, so that the description is omitted.
According to the method, the fourth association degree between each spectrum characteristic time sequence of the second remote sensing image of the sample point in the second time period and the preset spectrum characteristic time sequence of the corresponding type of the ground object type is calculated, the second association degree between the spectrum characteristic time sequence of the second remote sensing image of the sample point in the second time period and the preset spectrum characteristic time sequence of at least one ground object type is determined, further, the spectrum characteristic corresponding to the ground object type with the highest spectrum characteristic similarity of the annual change sample point in the second time period is determined, and finally the ground object type of the annual change sample point in the second time period is determined.
In the method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images, before the step of determining the second degree of association between the spectrum characteristic time sequence of the second remote sensing image in the second time period and the preset spectrum characteristic time sequence corresponding to at least one ground object type, the method further comprises the following steps:
determining a spectrum characteristic average value of a spectrum characteristic time sequence of a first remote sensing image of the sample point belonging to the same ground object type in a first time period at each moment;
integrating the spectrum characteristic average value at each moment belonging to the same feature type sample point into a spectrum characteristic time sequence of the corresponding feature type, and taking the spectrum characteristic time sequence as a preset spectrum characteristic time sequence of the feature type;
in the invention, the preset spectrum characteristic time sequence for determining the ground object type of the internationally-changed sample points in the second time period is obtained by calculating the spectrum characteristic time sequences of a plurality of sample points in the first time period.
Specifically, a spectral feature time sequence corresponding to each sample point in the target area in the first time period is obtained, and then the ground object type corresponding to each sample point in the target area is determined.
And counting the average value of the corresponding spectral characteristic values of the spectral characteristic time sequences of all sample points belonging to the same ground object type at each moment, integrating to obtain a preset spectral characteristic time sequence of the corresponding ground object type, and taking the preset spectral characteristic time sequence as a standard spectral characteristic time sequence of the ground object type.
And integrating the preset spectrum characteristic time sequence of each ground object type to obtain the preset spectrum characteristic time sequence corresponding to the at least one ground object type.
For example, if the object type of 300 sample points in the target area is a water body in the first time period, the average value of the spectrum characteristic values of each spectrum characteristic time sequence of the 300 sample points at each moment in the first time period is counted, the average value of the spectrum characteristic values at each moment is integrated, and a standard spectrum characteristic time sequence corresponding to the water body is generated and used as a preset spectrum characteristic value corresponding to the water body.
According to the method, the preset spectrum characteristic time sequence of each ground object type is obtained through calculation through the ground object type of each sample point in the first time period of the target area and the spectrum characteristic time sequence of the first remote sensing image of each sample point, and the preset spectrum characteristic time sequence is used for determining the ground object type of the internationally-changed sample point. In other words, when determining the standard spectrum characteristic time sequence corresponding to each ground object type, the influence of seasonal variation is considered, and then a more accurate preset spectrum characteristic time sequence is generated, so that the ground object type of the inter-year variation sample point is more accurately and automatically identified, meanwhile, the spectrum characteristic time sequence of the first remote sensing image in the first time period which is constructed is adopted in calculation, no additional sampling is needed, the steps for determining the preset spectrum characteristic value are simplified, and meanwhile, the accuracy and the robustness of the ground object sample automatic production method are improved.
After the preset spectrum characteristic time sequence is determined, the feature type identification of the annual change sample points of the target area in the second time period can be realized according to the preset spectrum characteristic time sequence, and then the feature type identification of all the sample points of the target area in the second time period is realized on the basis of determining the feature type of the annual stable sample points of the target area in the second time period according to the feature type of the annual stable sample points of the target area in the first time period, so that the production of the long-time sequence feature sample of the target area in the second time period is completed.
Further, reliability evaluation needs to be performed on the produced ground object sample type data.
Specifically, as shown in fig. 3, the ground object sample data are summarized and sorted, the identified annual stable samples and the identified annual change samples are combined, so that a ground object sample data set with a long time sequence is generated, and compared with the ground object sample data in the first time period, the number and the spatial distribution of the ground object samples in the second time period obtained by production are kept unchanged.
The reliability of the object samples of the object year is checked as accurately as possible, the object samples of the object region in the second time period are sampled in a layering and random manner, attention is required to be paid during sampling, and the proportion of the sampling points of all object types in the sampled samples is kept to be the same as the proportion of the object types of the original samples, so that the representativeness and the representativeness of the verification samples obtained by sampling are ensured.
The interpreter re-interprets the type of the ground object of the verification sample by visual interpretation. Specifically, the Google Earth high-resolution remote sensing image of the corresponding year is taken as a base chart, a verification sample is superimposed on the base chart, and the category attribute of each sample point in the verification sample is redetermined through visual interpretation of an interpreter.
Comparing the automatic identification result of the verification sample with the visual interpretation result, calculating a confusion matrix, and further evaluating the total Accuracy (OA), missing part Error (OE) and wrong part Error (CE) of the automatic identification result and the visual interpretation result, wherein the method is as follows:
（13）
（14）
（15）
in the method, in the process of the invention,diagonal matrix elements that are confusion matrices; />Is the->A row element sum representing an element sum of an automatic recognition category; />For confusion matrix->Column element sum, element sum characterizing visual interpretation category;Nto confuse the sum of all elements of the matrix. OA characterizes the overall accuracy of the sample identification,the larger the value, the higher the accuracy; OE and CE represent the missing division errors and the wrong division errors of sample identification, and the larger the numerical value is, the worse the accuracy is.
According to the invention, a sample migration algorithm based on a gray correlation model is constructed, manual intervention and threshold dynamic adjustment are not needed, so that labor cost and time cost are greatly reduced, ground object sample data production is automatically realized through computer algorithm modeling, the ground object type of sample points in a target time period can be automatically obtained through known ground object type data and spectrum data of remote sensing images in a corresponding time period, high-quality ground object sample data production with long time sequence and large area is effectively realized, and the method has high accuracy and universality and good application potential in the aspect of land coverage remote sensing drawing.
The invention relates to a method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images, which comprises the following steps before the step of determining the first association degree between the spectrum characteristic time sequence of a first remote sensing image of a sample point in a target area and the spectrum characteristic time sequence of a second remote sensing image of the sample point in a second time period:
after removing pixels which are shielded by cloud layers or cloud shadows in the remote sensing images at each moment in the first time period and the second time period, embedding and cutting the remote sensing images at each moment in the first time period and the second time period to obtain the first remote sensing image and the second remote sensing image, wherein the boundaries of the first remote sensing image and the second remote sensing image are consistent with the boundaries of the target areas to which the sample points belong.
When the remote sensing images of the target area in the first time period and the second time period are acquired, a plurality of remote sensing images of the area corresponding to the target area are acquired, and the target area is in the area range corresponding to the plurality of remote sensing images, and the acquired remote sensing images need to be preprocessed so as to acquire a first remote sensing image time sequence and a second remote sensing image time sequence of the target area in the first time period and the second time period.
Specifically, firstly, the acquired Landsat image is subjected to preprocessing such as radiation calibration, atmospheric correction, geometric correction and the like so as to reduce the geometric error and radiation error of the remote sensing image.
Further, pixels corresponding to portions of the target area in the remote sensing image that are blocked by cloud layers and/or cloud shadows need to be removed. Cloud mask processing is carried out by utilizing the quality wave band (QA_PIXEL wave band) of the Landsat satellite image, and invalid PIXELs corresponding to cloud layers and cloud shadows in the remote sensing image are removed.
On the basis, remote sensing images of the target area in each moment in the first time period and the second time period are determined.
Alternatively, the remote sensing image may be one or more images at each moment, so long as the target area is completely covered.
Specifically, according to the vector range of the target area, the remote sensing images at the same moment are subjected to mosaic processing to obtain the remote sensing images which completely cover the target area. And then taking the mosaic image as an input grid, taking the vector range of the target area as a cutting boundary, and finally obtaining the remote sensing images consistent with the target area boundary at each moment in the first time period and the second time period.
According to the invention, by preprocessing the remote sensing images in the first time period and the second time period in the target area, invalid pixels such as cloud layers and cloud shadows are removed, high-quality remote sensing images are obtained, the remote sensing images are inlaid and cut according to the vector range of the target area, and finally the first remote sensing images in the first time period and the second remote sensing images in the second time period of the target area are obtained, so that the method is used for producing ground object samples.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (4)
1. The automatic production method of the long-time-sequence ground object sample based on the dense time-sequence remote sensing image is characterized by comprising the following steps of:
determining a first association degree between a spectral feature time sequence of a first remote sensing image of a sample point in a target area in a first time period and a spectral feature time sequence of a second remote sensing image of the sample point in a second time period; generating a spectrum characteristic time sequence according to each spectrum characteristic value corresponding to the sample points in time sequence;
taking the ground object type of the sample point in a first time period as the ground object type of the sample point in a second time period under the condition that the first association degree is larger than a preset threshold value;
determining a second degree of correlation between the spectral feature time sequence of the second remote sensing image of the sample point in a second time period and a preset spectral feature time sequence of at least one ground object type under the condition that the first degree of correlation is smaller than or equal to the preset threshold;
taking the ground object type corresponding to the second association degree with the maximum degree as the ground object type of the sample point in the second time period;
the step of determining a first degree of correlation between the time series of the spectral features of the first remote sensing image of the sample point in the target area in the first time period and the time series of the spectral features of the second remote sensing image of the sample point in the second time period includes:
determining a third association degree between each spectrum characteristic time sequence of the sample point in the first remote sensing image of the first time period and the spectrum characteristic time sequence of the corresponding kind of the sample point in the second remote sensing image of the second time period;
determining the first association degree according to the third association degree;
the step of determining a second degree of association between the spectral feature time series of the second remote sensing image of the sample point in the second time period and the preset spectral feature time series of the at least one ground object type includes:
determining a fourth association degree between the spectrum characteristic time sequence of the sample point in the second remote sensing image of the second time period and the preset spectrum characteristic time sequence of the corresponding type of each ground object type;
and determining a second degree of correlation between the spectral feature time sequence of the second remote sensing image of the sample point in the second time period and a preset spectral feature time sequence corresponding to at least one ground object type according to the fourth degree of correlation.
2. The method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images according to claim 1, wherein the step of determining a second degree of correlation between the spectral feature time series of the second remote sensing image and the preset spectral feature time series of the at least one ground object type for the sample points in the second time period further comprises:
determining a spectrum characteristic average value of a spectrum characteristic time sequence of a first remote sensing image of the sample point belonging to the same ground object type in a first time period at each moment;
integrating the spectrum characteristic average value at each moment belonging to the same feature type sample point into a spectrum characteristic time sequence of the corresponding feature type, and taking the spectrum characteristic time sequence as a preset spectrum characteristic time sequence of the feature type;
and integrating the preset spectrum characteristic time sequence of each ground object type to obtain the preset spectrum characteristic time sequence corresponding to the at least one ground object type.
3. The method for automatically producing long-time-series ground object samples based on dense time-series remote sensing images according to claim 1 or 2, wherein the step of determining a first correlation between a spectral feature time sequence of a first remote sensing image of a first time period and a spectral feature time sequence of a second remote sensing image of a second time period of the sample points in the target area further comprises:
after removing pixels which are shielded by cloud layers or cloud shadows in the remote sensing images at each moment in the first time period and the second time period, embedding and cutting the remote sensing images at each moment in the first time period and the second time period to obtain the first remote sensing image and the second remote sensing image, wherein the boundaries of the first remote sensing image and the second remote sensing image are consistent with the boundaries of the target areas to which the sample points belong.
4. The method for automatically producing long-time-sequence ground object samples based on dense time-sequence remote sensing images according to claim 1 or 2, wherein the spectral features corresponding to the spectral feature time sequence comprise one or more of green band, red band, near infrared band, short-wave infrared 1, short-wave infrared 2, normalized vegetation index, normalized water index, improved normalized water index and automatic water extraction index of the remote sensing images.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311579977.9A CN117292154B (en) | 2023-11-24 | 2023-11-24 | Automatic production method of long-time-sequence ground object samples based on dense time-sequence remote sensing images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311579977.9A CN117292154B (en) | 2023-11-24 | 2023-11-24 | Automatic production method of long-time-sequence ground object samples based on dense time-sequence remote sensing images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117292154A CN117292154A (en) | 2023-12-26 |
| CN117292154B true CN117292154B (en) | 2024-02-06 |
Family
ID=89239375
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311579977.9A Active CN117292154B (en) | 2023-11-24 | 2023-11-24 | Automatic production method of long-time-sequence ground object samples based on dense time-sequence remote sensing images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117292154B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114202705A (en) * | 2022-02-16 | 2022-03-18 | 清华大学 | Spectral feature time sequence construction method and system |
| CN116129284A (en) * | 2022-12-22 | 2023-05-16 | 农业农村部规划设计研究院 | Remote sensing extraction method for abandoned land based on time sequence change characteristics |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110472525B (en) * | 2019-07-26 | 2021-05-07 | 浙江工业大学 | Noise detection method for time series remote sensing vegetation index |
| CN114419463B (en) * | 2022-01-26 | 2022-09-30 | 河南大学 | Cloud platform-based global solar photovoltaic panel remote sensing automatic identification method |
-
2023
- 2023-11-24 CN CN202311579977.9A patent/CN117292154B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114202705A (en) * | 2022-02-16 | 2022-03-18 | 清华大学 | Spectral feature time sequence construction method and system |
| CN116129284A (en) * | 2022-12-22 | 2023-05-16 | 农业农村部规划设计研究院 | Remote sensing extraction method for abandoned land based on time sequence change characteristics |
Non-Patent Citations (2)
| Title |
|---|
| Wetland Mapping and Landscape Analysis for Supporting International Wetland Cities: Case Studies in Nanchang City and Wuhan City;Geng Zhipeng;《IEEE》;8858-8870 * |
| 水体透明度遥感反演算法研究进展;赵春燕;《研究综述》;176-185 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117292154A (en) | 2023-12-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Du et al. | Radiometric normalization, compositing, and quality control for satellite high resolution image mosaics over large areas | |
| CN112712535B (en) | Mask-RCNN landslide segmentation method based on simulation difficult sample | |
| CN108195767B (en) | Estuary wetland foreign species monitoring method | |
| CN112541921B (en) | Urban green land vegetation information data accurate determination method | |
| CN110120046B (en) | Potential landslide identification method integrating DEM (digital elevation model), optical remote sensing and deformation information | |
| CN111738144A (en) | Surface water product generation method and system based on Google Earth Engine cloud platform | |
| CN109767409B (en) | Landslide change detection method based on remote sensing image, storage medium and electronic equipment | |
| CN111008664B (en) | Hyperspectral sea ice detection method based on space-spectrum combined characteristics | |
| CN111047566A (en) | Method for carrying out aquatic vegetation annual change statistics by unmanned aerial vehicle and multispectral satellite image | |
| CN107688776B (en) | Urban water body extraction method | |
| Zhang et al. | A mixed radiometric normalization method for mosaicking of high-resolution satellite imagery | |
| CN114419463B (en) | Cloud platform-based global solar photovoltaic panel remote sensing automatic identification method | |
| CN115546656A (en) | Remote sensing image breeding area extraction method based on deep learning | |
| CN115453555A (en) | Unmanned aerial vehicle rapid monitoring method and system for grassland productivity | |
| CN113936226B (en) | Global glacier search identification method based on remote sensing cloud computing | |
| Bin et al. | Parcel-based change detection in land-use maps by adopting the holistic feature | |
| CN117292154B (en) | Automatic production method of long-time-sequence ground object samples based on dense time-sequence remote sensing images | |
| Sefercik et al. | 3D positioning accuracy and land cover classification performance of multispectral RTK UAVs | |
| CN117218201A (en) | Unmanned aerial vehicle image positioning precision improving method and system under GNSS refusing condition | |
| Gupta et al. | Deep learning based detection of water bodies using satellite images | |
| Misra et al. | An approach for generation of multi temporal co-registered optical remote sensing images from Resourcesat-2/2A sensors | |
| CN115791639A (en) | Soybean planting area remote sensing extraction method based on phenological information | |
| CN115082812A (en) | Agricultural landscape non-agricultural habitat green patch extraction method and related equipment thereof | |
| CN114266968B (en) | Remote sensing automatic interpretation method for different land coverage types of city | |
| CN116883853B (en) | Crop space-time information remote sensing classification method based on transfer learning |
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 |