Forest type remote sensing classification method based on historical sample migration

Abstract

The invention provides a forest type remote sensing classification method based on historical sample migration, and belongs to the field of forest monitoring. The method comprises the steps of firstly determining a monitoring area, a historical year and a target year, obtaining corresponding sample points, then obtaining long-time-sequence multi-source remote sensing data of the historical sample points and representing forest growth succession features, then judging whether a forest type is replaced or not, and migrating sample data which is not replaced to the target year; extracting classification features related to remote sensing classification by using the multi-source remote sensing data of the target year; dividing the monitoring area into a plurality of plant regions, and selecting the optimal classification feature of each plant region from the related classification features; finally, based on the optimal classification features and migrated historical forestry field survey sample point data which are not replaced, forest types of all plant regions are classified. According to the invention, the number of samples in the target year is increased, and the accuracy of mountain forest type remote sensing classification under the small sample condition is improved.

Description

Technical Field

[0001] The invention belongs to the field of forest monitoring, and in particular relates to a forest type remote sensing classification method based on historical sample migration. Background Art

[0002] Mountain forests are rich in variety and are the main ecological security barrier, an important reservoir of natural resources and a treasure trove of biodiversity. Accurate classification and mapping of mountain forest types is of great scientific significance for sustainable management of forest resources, biodiversity conservation and ecosystem service evaluation. Traditional mountain forest type classification relies on field survey data; with the development of satellite remote sensing technology, it has become the main means of mapping mountain forest types due to its advantages such as wide coverage, short revisit cycle and low data acquisition cost.

[0003] In recent years, remote sensing technology has been used to classify mountain forests, generally using traditional machine learning and deep learning methods. However, the classification accuracy of such methods is heavily dependent on the quality and quantity of real samples. Mountainous areas have complex terrain conditions, changeable environmental factors, and low traffic accessibility, which often makes it difficult to obtain a large number of forestry field survey samples. As a result, the above methods generally have problems such as low classification accuracy and large spatial differences in credibility in remote sensing classification of mountain forest types.

[0004] In the prior art, the number of samples is generally expanded by migrating historical forestry field survey samples to the target year, thereby improving the accuracy of remote sensing classification. The national forestry field survey data over the years contains a large number of historical known samples. If these sample information can be migrated to the target year to expand the number of samples, it will fundamentally solve the small sample problem faced in the remote sensing classification of mountain forest types. However, the existing sample migration method is not yet mature, with low migration efficiency and poor quality, and still cannot effectively improve the precision and accuracy of remote sensing classification of forest types. Summary of the invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, the present invention aims to provide a remote sensing classification method for forest types based on historical sample migration, which utilizes the growth succession characteristics from the historical year to the target year to migrate the historical forestry field survey sample point data where no forest type change has occurred to the target year, so as to expand the number of samples in the target year, and combine the overall sample after sample expansion with the classification of flora to carry out remote sensing classification of forest types, so as to improve the classification precision and accuracy of forest types under small sample conditions.

[0006] In order to achieve the above purpose, the embodiment of the present invention adopts the following technical solution:

[0007] A remote sensing classification method for forest types based on historical sample migration, the method comprising the following steps:

[0008] Step S1, determining the monitoring area, historical years and target years, obtaining historical sample points of forestry field surveys in historical years and target sample points of forestry field surveys in the target year within the monitoring area;

[0009] Step S2, obtaining long-term multi-source remote sensing data of the historical sample points from the historical year to the target year;

[0010] Step S3, using long-term multi-source remote sensing data to characterize the forest growth succession characteristics of historical sample points;

[0011] Step S4, based on the forest growth succession characteristics, determine whether the forest type of the historical sample point has changed from the historical year to the target year; if the forest type label of the target year is the same as the forest type label of the historical year, it is determined that no change has occurred; if the forest type label of the target year is different from the forest history label of the historical year, it is determined that a change has occurred;

[0012] Step S5, according to whether the forest type has changed, the forestry field survey sample data of the historical sample points that have not changed are migrated to the target year, and the historical sample points that have changed are deleted;

[0013] Step S6, extracting classification features related to remote sensing classification of regional forest types using multi-source remote sensing data of the target year;

[0014] Step S7, dividing the monitoring area into a plurality of floras according to the spatial distribution characteristics of the floras in the monitoring area, and selecting the optimal classification features for forest type classification of each flora from the relevant classification features using an association hierarchical clustering method;

[0015] Step S8, based on the optimal classification features and the forestry field survey sample data of the historical sample points that have not been replaced and migrated to the target year, use the random forest algorithm to classify the forest type of each flora;

[0016] Step S9, integrating the classification results of forest type classification of all floras to obtain a spatial distribution map of forest types in the entire monitoring area.

[0017] As a preferred embodiment of the present invention, step S1 describes obtaining forestry field survey sample point data for historical years and target years in the monitoring area. First, forestry field survey data for historical years and target years in the monitoring area are obtained, and then the surface-to-point method is used to generate historical sample point data and target sample point data maps, respectively.

[0018] As a preferred embodiment of the present invention, the long-term multi-source remote sensing data in step S2 includes all available Landsat Collection2Tier1 surface reflectance data images, Sentinel-1 / 2 data, SRTM-DEM data, climate data and soil moisture data on the Google Earth Engine GEE platform corresponding to the period from historical years to target years.

[0019] As a preferred embodiment of the present invention, the forest growth succession characteristics in step S4 include the spectral similarity of historical sample points in historical years and target years, the time series trajectory of historical sample points from historical years to target years, and the environmental similarity of historical sample points and target sample points of the same type.

[0020] As a preferred embodiment of the present invention, step S4 determines whether the forest type of the historical sample point has changed from the historical year to the target year by using a threshold segmentation method.

[0021] As a preferred embodiment of the present invention, the threshold segmentation formula is as follows:

[0022]

[0023] And, Environmental similarity=max(S i1 ,S i2 ,…,S ij )

[0024] In formula (8), Foresttype is the forest type label of the historical sample point; SAD and ED are spectral similarity indicators that quantify the historical sample point between the historical year and the target year, σ1 and σ2 are the segmentation thresholds of SAD and ED respectively; disturbance year is the forest disturbance detection result of the historical sample point from the historical year to the target year, Historical year is the historical year; Environmental similarity is the environmental similarity between the historical sample point and the target sample point of the same type, σ3 is the segmentation threshold of environmental similarity, S i,j is the environmental similarity between all sample points of the same forest type in the historical sample point i and the target sample point j.

[0025] As a preferred embodiment of the present invention, the value of σ1 is 0.95; the value of σ2 is 0.1.

[0026] As a preferred embodiment of the present invention, the migration model of the forestry field survey sample data of the historical sample points in step S5 is:

[0027]

[0028] In formula (9), the forest type historical sample point is the forest type of the historical sample point in each historical forestry field survey sample from the historical year to the target year. If the forest type of the historical sample point has not changed, it means that the forest type label of the sample point has not changed, and it is migrated to the target year for forest type remote sensing classification; otherwise, the type label information of the sample point has changed, and the sample point is deleted.

[0029] As a preferred embodiment of the present invention, the relevant classification features extracted in step S6 include: spectrum, polarization, texture, phenology and environment.

[0030] As a preferred embodiment of the present invention, the method further includes:

[0031] Step S10, evaluating mapping accuracy using forestry field survey sample data of the target year.

[0032] The technical solution provided by the embodiment of the present invention has the following beneficial effects:

[0033] The present invention provides a forest type remote sensing classification method based on historical sample migration. Based on the growth succession information of historical samples from the historical year to the target year, historical samples where forest type replacement has not occurred are migrated to the target year to achieve remote sensing classification of forest types in the target year. It is well known that even without interference from natural or human factors, forest communities will continue to renew, grow and succeed over time; this growth succession process is usually accompanied by changes in characteristics such as tree species composition, forest stand structure, canopy coverage and vegetation functional traits, and these characteristic changes often cause changes in forest canopy spectral characteristics and local microclimate. For example, in the pioneer stage (such as after a forest fire or after the retreat of glaciers), forest stands are mainly dominated by fast-growing tree species (such as poplars and birches); in the secondary forest and mature forest stages, they are gradually replaced by taller shade-tolerant tree species (such as oaks and maples), gradually forming a stable forest stand structure and forest ecosystem; the overmature forest stage may contain endangered or special environmental adapted tree species, and the forest stand structure becomes more complex and diversified. The present invention overcomes the problem that the existing methods ignore the influence of forest growth succession on sample migration quality, and can effectively improve the migration quality of historical forestry field survey samples and improve the classification accuracy of forest types in the target year.

[0034] Of course, it is not necessary to achieve all of the advantages described above at the same time to implement any product or method of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 It is a flow chart of a forest type remote sensing classification method based on historical sample migration provided by an embodiment of the present invention;

[0037] Figure 2 is a spatial distribution pattern diagram of forest types at altitude in the monitoring area in 2023, which is inferred from a specific application example of an embodiment of the present invention;

[0038] Figure 3 This is a comparison chart of PA indicators of a specific application example of an embodiment of the present invention and other existing methods for remote sensing classification of forest types;

[0039] Figure 4 It is a comparison chart of UA indicators of a specific application example of an embodiment of the present invention and other existing methods for remote sensing classification of forest types;

[0040] Figure 5 This is a comparison chart of F1 scores of a specific application example of an embodiment of the present invention and other existing methods for remote sensing classification of forest types;

[0041] Figure 6 This is a comparison chart of the number of migrated samples of different forest types using a specific application example of an embodiment of the present invention and other existing methods for migrating historical samples. DETAILED DESCRIPTION

[0042] After discovering the above problems, the inventors of this application conducted a detailed study on the existing remote sensing classification methods for forest types, especially the methods using historical migration samples for classification. The study found that there are currently several methods for migrating historical field survey samples for target year surface cover classification when classifying forest types by remote sensing. In summary, they mainly include historical sample reuse method, historical sample weighting method and dual-temporal image change detection method, but all of them have certain shortcomings.

[0043] For example, the historical sample weighting method dynamically adjusts the sample weights based on the contribution of historical samples to the target model, and uses the weighted historical samples to classify the land cover of the target year. For example, Zhou et al. (2016) introduced a transfer term into the target classification model, and iteratively adjusted the sample weights based on the classification accuracy of historical and target samples in the model (if the historical sample is misclassified, it indicates that the sample is inconsistent with the spatial distribution of the remote sensing data of the target year, and its weight should be reduced; if the target sample is misclassified, it indicates that the sample has not been fully learned by the model, and its weight should be increased) until all target samples are correctly classified, and the weighted samples are used for land cover classification. Compared with the historical sample reuse method, although it reduces the impact of historical samples that are inconsistent with the spatial distribution of remote sensing data of the target year on the classification accuracy, it does not take into account the potential impact of the replacement of historical sample label information caused by land cover changes on the accuracy of land cover classification in the target year.

[0044] Another example is the dual-temporal image change detection method, which assumes that changes in surface cover types will cause changes in surface radiation characteristics; first, the spectral feature similarity between the historical and target year remote sensing data is calculated, and then a similarity threshold is set to determine whether the label information of the historical samples has changed, and the historical samples where the label information has not changed are migrated to the target year; finally, based on the migrated samples and the remote sensing data of the target year, a machine learning algorithm is used to achieve the surface cover classification of the target year. For example, Lin et al. (2019) migrated historical samples with no label information change to the target year by comparing the spectral feature differences (Change Vector Analysis, CVA) between historical samples (2010) and the target year, and used the migrated samples and Landsat data to realize automatic classification of surface cover types in rapidly urbanized areas; for another example, Huang et al. (2020) based on the global surface cover type sample dataset (Firstall-season sample set) collected by the FROM-GLC project in 2015, calculated the spectral similarity and spectral distance between the target year and the 2015 Landsat data, migrated historical samples with no label information change to the target year, and used these migrated samples and Landsat data to classify global surface cover in 1990, 1995, 2000, 2005 and 2010 respectively. Although this type of method reduces the impact of the replacement of historical sample label information on the quality of migrated samples to a certain extent, it still has the following two limitations in the migration process of historical forestry field survey samples: 1) The spectral characteristics of different forest types in multispectral remote sensing images are relatively small. Only using dual-phase remote sensing images can usually only identify the replacement of forests and non-forests, and it is difficult to effectively identify the replacement information of forest types at the stand scale, such as historical samples from coniferous / broad-leaved forests to mixed coniferous and broad-leaved forests; 2) This method does not fully consider the potential impact of the natural growth succession of forests from historical years to target years on the quality of sample migration.

[0045] The above analysis shows that the historical sample weighting method is cumbersome, and it requires a cumbersome dynamic weight adjustment process to generate samples suitable for the target year's land cover classification. Although the dual-temporal remote sensing image change detection method has been improved on the basis of historical sample reuse, these methods do not consider the impact of the natural growth succession of the forest from the historical year to the target year on the quality of sample migration, resulting in low quality of migrated forestry field survey samples. It is well known that even in the absence of natural or human interference, forest communities will continue to renew, grow and succeed over time; this growth succession process is usually accompanied by changes in characteristics such as tree species composition, stand structure, canopy coverage and vegetation functional traits, and these characteristic changes often cause changes in forest canopy spectral characteristics and local microclimate. For example, in the pioneer stage (such as after a forest fire or after the retreat of glaciers), forest stands are dominated by fast-growing tree species (such as poplars and birches); in the secondary forest and mature forest stages, they are gradually replaced by taller shade-tolerant tree species (such as oaks and maples), gradually forming a stable forest structure and forest ecosystem; the overmature forest stage may contain endangered or special environmentally adapted tree species, and the forest stand structure becomes more complex and diverse. Therefore, in the process of migrating historical forestry field survey samples, it is necessary to comprehensively consider the impact of forest growth succession on the quality of sample migration, so as to improve the quality of migrated samples, increase the number of effective samples in the target year, and improve the accuracy of remote sensing classification of mountain forest types under small sample conditions. However, the monitoring methods of existing technologies do not take the above issues into account.

[0046] It should be noted that the defects existing in the solutions in the above-mentioned prior art are the results obtained by the inventor after practice and careful research. Therefore, the discovery process of the above-mentioned problems and the solutions proposed in the embodiments of the present invention for the above-mentioned problems below should all be the contributions made by the inventor to the present invention in the process of the present invention.

[0047] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can be arranged and designed in various different configurations. It should be noted that the embodiments of the present invention and the features in the embodiments can also be combined with each other without conflict.

[0048] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0049] After the above in-depth analysis, the embodiment of the present invention provides a remote sensing classification method for forest types based on historical sample migration. The growth succession information of historical samples from the historical year to the target year is used to migrate the historical samples in the same monitoring area where no forest type replacement has occurred to the target year, so as to achieve remote sensing classification of forest types in the target year. First, multi-source remote sensing data from the historical year to the target year are used to quantitatively characterize the growth succession characteristics of the forest; then, the growth succession characteristics are used to determine whether the forest type labels of the historical samples have been replaced, and the historical samples that have not been replaced are migrated to the target year; finally, based on the multi-source remote sensing data and the migrated samples, a machine learning algorithm is used to perform remote sensing classification of forest types in the target year, and the forestry field survey data of the target year is used to quantitatively evaluate the effectiveness of the automatic migration method of historical forestry field survey samples in remote sensing classification of mountain forest types.

[0050] like Figure 1 As shown, the forest type remote sensing classification method based on historical sample migration includes the following steps:

[0051] Step S1, determine the monitoring area, historical years and target years, obtain historical sample points of forestry field surveys in historical years and target sample points of forestry field surveys in the target year within the monitoring area.

[0052] In this step, the forestry field survey sample point data of the historical years and the target years in the monitoring area are obtained. First, the forestry field survey data of the monitoring area in the historical years and the target years are obtained, and then the historical sample point data and the target sample point data are generated respectively using the surface-to-point method.

[0053] Step S2, obtaining long-term multi-source remote sensing data of the historical sample points from the historical year to the target year.

[0054] In this step, the long-term multi-source remote sensing data include all available Landsat Collection 2Tier1 surface reflectance data images, Sentinel-1 / 2 data, SRTM-DEM data, climate data and soil moisture data on the Google Earth Engine GEE platform corresponding to the period from historical years to target years.

[0055] Step S3, using long-term multi-source remote sensing data to characterize the forest growth succession characteristics of historical sample points; the forest growth succession characteristics include the spectral similarity of historical sample points in historical years and target years, the time series trajectory of historical sample points from historical years to target years, and the environmental similarity of historical sample points and target sample points of the same type.

[0056] In this step, the spectral similarity between the historical sample points in the historical year and the target year can be calculated by using methods such as spectral angular distance SAD and Euclidean distance ED; the time series trajectory of the historical sample points from the historical year to the target year can be calculated by using the improved LandTrendr algorithm iLandTrendr; the environmental similarity between the historical sample points and the same type of target sample points can be calculated by using the random forest regression algorithm, the relationship between the Gaussian similarity function and various geographic environmental covariates, and other methods.

[0057] In a specific embodiment, the spectral similarity between historical sample points in historical years and target years is calculated, Landsat data of the monitoring area in historical years and target years are collected, and the spectral angular distance (SAD) and Euclidean distance (ED) are used to quantify the spectral similarity between the optical remote sensing images of historical samples in historical years and target years. Specifically, the following steps are included:

[0058] Step S311, collecting Landsat data of historical sample points corresponding to historical years and target years;

[0059] Step S312, using SAD and ED to calculate the spectral similarity between the historical sample points in the historical year and the target year Landsat image, the calculation formula is as follows:

[0060]

[0061]

[0062] Where θ is the spectral angle, is the reference spectral feature corresponding to the sample point t1 in historical year, is the spectral feature corresponding to the historical sample in the target year t2; i is the corresponding spectral band, which represents the B1-B5 and B7 bands of Landsat5TM data or the B2-B7 bands in Landsat8 OLI; if the spectral feature does not change between the historical and target years, SAD is equal to 1 and ED is equal to 0. Preferably, the segmentation thresholds of SAD and ED are set to 0.95 and 0.1, respectively.

[0063] In another specific embodiment, the time series trajectory of the historical sample points from the historical year to the target year is calculated, and the Landsat data of the vegetation growth season each year from the historical year to the target year is collected, and then the area covered by clouds and cloud shadows is masked, and then the masked image is synthesized to generate an image that can well capture the Landsat observation value, and finally based on the long-term Landsat data, the iLandTrendr (Improved Landsat-based detection of trends in disturbance and recovery) algorithm is used to detect whether the historical sample has experienced forest replacement during this period, that is, to detect whether the historical sample data has experienced forest disturbance events such as forest fires, deforestation, pests and diseases. Specifically, the following steps are included:

[0064] Step S321, collecting Landsat data of vegetation growing seasons in each year within a long time series, masking the areas covered by clouds and cloud shadows, and then synthesizing the masked images to generate an image that can well capture Landsat observations;

[0065] Step S322, calculate the normalized burn index NBR of each pixel, then use linear interpolation to fill in the NBR value corresponding to the year covered by clouds and cloud shadows pixel by pixel, and finally use the SG filtering algorithm to filter out data noise to obtain the NBR time series trajectory of each pixel changing over time;

[0066] Step S323, based on the NBR time series trajectory within the long time series year range, use the iLandTrendr algorithm to detect the year when forest disturbance occurs in each pixel; that is, assuming that when forest disturbance occurs, the NBR time series trajectory will show a decreasing value, and when the forest recovers, the NBR index will show a gradually increasing trend. Based on the iLandTrendr algorithm, the year corresponding to the node where the NBR value "decreases-increases" changes occur is considered to be the year when forest disturbance occurs, and the corresponding pixel is regarded as the pixel where forest disturbance occurs within the long time series year range.

[0067] In another specific embodiment, the calculation of the environmental similarity between the historical sample points and the target sample points of the same type firstly uses the multi-source remote sensing data of the target year to characterize the environmental conditions related to the spatial distribution of forest types, such as forest canopy characteristics (spectrum, polarization, phenology, spatial texture, etc.) and geographical environment (precipitation, temperature, soil, etc.); secondly, uses the Gaussian similarity function to compare the similarity of the environmental conditions between the historical sample points and the target sample points, specifically including the following steps:

[0068] Step S331, using multi-source remote sensing data acquired in the target year to comprehensively characterize the environmental conditions of each geographical location;

[0069] Step S332, calculate the environmental similarity between the historical sample point and the target sample point, and the calculation formula is as follows:

[0070]

[0071] In formula (4), and are the values ​​of historical sample point i and target sample point j on the vth environmental covariate, respectively; m is the number of geographic environmental covariates used to characterize the spatial distribution of forest types, which is based on the forest type samples of the target year, using the random forest algorithm to evaluate the importance of all classification features in step S2, and select features with feature importance greater than 0.00001 to characterize the geographic environment configuration corresponding to each sample point. v (·) is a function used to calculate the geographical environment similarity between historical sample point i and target sample point j on a single environmental variable v. This paper uses the Gaussian similarity function for calculation, and its calculation formula is shown in Formula 7:

[0072]

[0073] In formula (5), is the standard deviation of the vth environmental covariate in the whole region, It is the square root of the average deviation of the j-th target sample point to all historical sample points of the same type (i = 1, 2, ..., k) on the v-th environmental covariate:

[0074]

[0075] It is a function used to synthesize the similarity of m environmental covariates. The similarity of each environmental covariate is synthesized by using the characteristic importance of each environmental covariate to the forest type classification to obtain the comprehensive environmental similarity between the historical sample point i and the target sample point j:

[0076]

[0077] In formula (7), The Mean Decrease Accuracy (MDA) method in the random forest algorithm is used to evaluate the importance of all features and the feature importance of each environmental covariate to the classification of regional forest types.

[0078] Step S4, based on the forest growth succession characteristics, determine whether the forest type of the historical sample point has changed from the historical year to the target year; if the forest type label of the target year is the same as the forest type label of the historical year, it is determined that no change has occurred; if the forest type label of the target year is different from the forest history label of the historical year, it is determined that a change has occurred.

[0079] In this step, the threshold segmentation method is used to determine whether the forest type of the historical sample point has changed from the historical year to the target year. Specifically, the threshold segmentation formula is as follows:

[0080]

[0081] And, Environmental similarity=max(S i1 ,S i2 ,…,S ij )

[0082] In formula (8), Foresttype is the forest type label of the historical sample point; SAD and ED are spectral similarity indicators that quantify the historical sample point between the historical year and the target year; σ1 and σ2 are the segmentation thresholds of SAD and ED respectively; the smaller the ED value or the larger the SAD value, the more similar the spectrum is, and the smaller the possibility of forest type replacement; disturbance year is the forest disturbance detection result of the historical sample point from the historical year to the target year. If the year of forest disturbance is less than or equal to the historical year, it means that there is no forest disturbance at the sample location, and the possibility of forest type label information replacement is small; environmental similarity is the environmental similarity between the historical sample point and the target sample point of the same type; σ3 is the segmentation threshold of environmental similarity, S i,j is the environmental similarity between all sample points of the same forest type in the historical sample point i and the target sample point j. Based on the third law of geography, the present invention assumes that similar environmental configurations have similar forest types, that is, the greater the environmental similarity between the historical sample point and the target sample point, the smaller the possibility of forest type replacement; therefore, the present invention uses each historical sample point collected in the forestry field in the target year as a case containing a specific "forest type-environmental condition" relationship, which can represent the appropriate geographical environmental conditions and forest canopy characteristics of the forest type.

[0083] Step S5, according to whether the forest type has changed, the forestry field survey sample data of the historical sample points that have not changed are migrated to the target year, and the historical sample points that have changed are deleted.

[0084] In this step, the migration model of forestry field survey sample data of historical sample points is:

[0085]

[0086] In formula (9), the forest type historical sample point is the forest type of the historical sample point in each historical forestry field survey sample from the historical year to the target year. If the forest type of the historical sample point has not changed, it means that the forest type label of the sample point has not changed and can be migrated to the target year for forest type remote sensing classification; otherwise, the type label information of the sample point has changed, and the sample point is deleted.

[0087] Step S6, using the multi-source remote sensing data of the target year to extract classification features related to the remote sensing classification of regional forest types; as shown in Table 1, the extracted relevant classification features include: spectrum, polarization, texture, phenology and environment.

[0088] Table 1. Forest type classification feature extraction

[0089]

[0090] The spectral features include: 10 bands of Sentinel-2, NDVI, SVVI, EVI, NDRE, REPI, REI, MCARI, NDPI, MCR, IRECI, NDVIR1, TC_Greenness, etc. The polarization features include: VV, VH, ratio, DIF, AVE, NDI, RVI, mRVI, VDDPI; the texture features include: Asm, contract, dvar... (Calculation for red_edge1 using GLCM); the phenological features include: REP (std, mean, median, min, max, Q25, Q75, IQR, cv, zf), std (mRVI, VV, VH), mRVI_summerwinter, NDVI_maxsummer; the environmental features include: Elevation, Slope, Aspect, PrecMean, PrecMs, TemMean, TemMs, SMRI.

[0091] Step S7, according to the spatial distribution characteristics of the flora in the monitoring area, the monitoring area is divided into multiple floras, and the optimal classification features for forest type classification of each flora are selected from the relevant classification features using the associated hierarchical clustering method to reduce the impact of collinearity and redundancy between classification features on the classification results.

[0092] Step S8, based on the optimal classification features and the forestry field survey sample data of the historical sample points that have not been replaced and migrated to the target year, use the random forest algorithm to classify the forest type of each flora.

[0093] Step S9, integrating the classification results of forest type classification of all floras to obtain a spatial distribution map of forest types in the entire monitoring area.

[0094] This embodiment may also include:

[0095] Step S10, evaluating mapping accuracy using forestry field survey sample data of the target year.

[0096] Yunnan Province in southwestern China is taken as the study area. The forest type remote sensing classification method based on historical sample migration provided by the embodiment of the present invention is used to perform remote sensing classification on the forest types in the selected area. The selected historical years are 2012 and 2016, and the target year is 2023.

[0097] First, the historical forestry field survey samples from 2012 and 2016 were migrated; then, based on the migrated samples, remote sensing classification of forest types in 2023 was carried out.

[0098] The monitoring area is located in southwest China, east of the Himalayas, adjacent to Myanmar, Vietnam and Laos. It is an important part of the India-Myanmar biodiversity hotspot. The forest coverage rate is as high as 65% and the forest stands are relatively young. The forest growth is vigorous and the forest carbon sink potential is large. The altitude of the study area is between 76.4 and 6740m, and it descends in a stepped manner from north to south. The landform types are complex and diverse, including plains, mountains, hills, basins, etc. The study area has a subtropical plateau monsoon climate, but due to its complex terrain conditions, the regional temperature and precipitation conditions have significant spatial heterogeneity, and the spatial and vertical differences of local climate are large, including various ecosystems except oceans and deserts. The unique and diverse natural environmental conditions in the monitoring area provide suitable habitats for the origin, evolution and reproduction of various organisms. It contains rich forest resources and contains a variety of flora. The eastern part is mainly Sino-Japanese flora, the western part is Sino-Himalayan flora, and the southern part is mainly distributed with paleotropical flora. Accurate forest type classification can provide important basic data for the sustainable management of regional forest resources.

[0099] In this embodiment, forest type sample point data within the monitoring area are collected, including Yunnan Province forestry field survey data for historical years and target years.

[0100] Among them, the historical sample data includes forestry field survey data from 2012 and 2016 (3120), of which there are 3699 samples in 2012 and 14065 samples in 2016; the forestry field survey samples in the target year are 3210. In this embodiment, the acquisition of forest type samples of each sample plot is completed in 4 steps: 1) a series of sample plots are laid out in Yunnan Province based on comprehensive consideration of spatial representativeness, economy and traffic accessibility; 2) according to the preset longitude and latitude of the sample plot, a 30×30m forest sample plot is laid out using a leather rope, and real-time dynamic differential positioning technology (Real Time KinematicGPS, RTK-GPS) is used to dynamically adjust the actual longitude and latitude information of the sample plot center and four corner points to determine the sample plot boundary; 3) for each sample plot, the diameter at breast height of each tree is measured using a breast diameter ruler, and the forest type to which each tree species with a breast diameter of ≥5cm belongs is manually identified according to the shape of the leaves, and its crown coverage area is quantitatively estimated; 4) the forest type to which the sample plot belongs is determined according to the proportion of crown coverage area of ​​each forest type. If the proportion of coniferous forest and broad-leaved forest is less than 65%, the sample plot belongs to a mixed coniferous and broad-leaved forest, otherwise it is a pure forest.

[0101] Then obtain the long-term multi-source remote sensing data of the historical sample points from the historical year to the target year. The long-term multi-source remote sensing data includes all available Landsat Collection 2Tier1 surface reflectance data images, Sentinel-1 / 2 data, SRTM-DEM data, climate data and soil moisture data on the Google Earth Engine GEE platform corresponding to the period from the historical forest survey year (2012 / 2016) to the target year (2023).

[0102] After corresponding preprocessing of the above long-term multi-source remote sensing data, the forest growth succession characteristics of the historical sample points are characterized.

[0103] The description of forest growth succession characteristics in this step is mainly completed in three steps, including:

[0104] The spectral angular distance SAD and the Euclidean distance ED are used to quantify the spectral similarity between historical samples in historical years and target years; the spectral similarity includes the similarity between the B1-B7 bands contained in the Landsat data of historical years (2012 / 2016) and target years (2023);

[0105] The Landsat data from historical samples between the historical year and the target year were used to quantify the time series trajectory, and the iLandTrendr algorithm was used to detect whether forest disturbance occurred during this period;

[0106] The environmental similarity between historical sample points in the region and the same type of sample points in each target year was calculated using the Gaussian similarity function and the characteristic importance of each geographic environmental covariate.

[0107] Based on the characteristics of forest growth succession, the forestry field survey sample data corresponding to the historical sample points judged as having not undergone forest type replacement will be migrated to the forestry field survey sample data of 2023 to complete the migration of historical data.

[0108] Based on the migrated data and the multi-source remote sensing data in 2023, a machine learning algorithm is used to perform remote sensing classification of forest types in the target year, and steps S6-S10 are executed.

[0109] According to the spatial distribution characteristics of Yunnan's flora, Yunnan Province is divided into multiple floras, and the association hierarchical clustering method is used to select the preferred features that are beneficial to the classification of forest types in each plant region from the classification feature set shown in Table 1 to reduce the impact of collinearity and redundancy between classification features on the classification results. The classification features mainly include spectral, polarization, phenological, texture and geographical environment characteristics; the preferred classification features of each flora are shown in Table 2:

[0110] Table 2 Preferred classification characteristics of each flora

[0111]

[0112]

[0113] Figure 2 The spatial distribution map of forest types in the entire monitoring area is obtained by integrating the classification results of 8 floras.

[0114] In order to verify the effectiveness of the method of automatic migration of historical forestry field survey samples based on remote sensing forest classification (a samples transferring method considering forest succession, STM-CFS) proposed in this paper, this example uses the forestry field survey samples of the target year to quantitatively evaluate the mapping accuracy of the time series forest canopy height distribution map inferred by this method. The quantitative evaluation indicators include overall accuracy (OA), Kappa coefficient, producer's accuracy (PA), user's accuracy (UA) and F1 score (F1score), and the calculation formula is as follows:

[0115]

[0116]

[0117]

[0118]

[0119]

[0120] Where n is the forest type, N is the total number of forest type samples surveyed in 2023, and X is ii represents the elements on the diagonal of the confusion matrix, that is, the total number of correctly classified pixels in each forest type, X i+ and X +i They represent the column sum and row sum of each forest type, that is, the total number of misclassified and misclassified pixels.

[0121] The classification accuracy of the forest type classification map of the monitoring area in 2023 inferred by the method of this embodiment is compared with the field forestry survey sample point data in 2023, and the results are as follows:

[0122] As can be seen from Table 3, the quality of training samples is positively correlated with classification accuracy. Compared with directly applying all historical samples to the forest type classification of the target year (OA: 64.94%, kappa: 0.4059), the dual-temporal image change detection method achieved higher classification accuracy in the remote sensing classification of mountain forest types (OA: 73.27%, kappa: 0.5290). The main reason for this phenomenon may be the widespread existence of natural forest growth succession, which leads to a certain proportion of historical samples that have changed from forest to non-forest in the target year, that is, false positive samples; the dual-temporal image change detection algorithm effectively filters out such samples, improves the quality of training samples to a certain extent, and improves the classification accuracy of forest types. However, the dual-temporal image change detection algorithm can only detect samples with significant spectral feature changes, ignoring the impact of spectral feature changes caused by the growth and replacement of forest types on the quality of sample migration, and its migrated forest type samples still contain a large number of false positive samples. Compared with the change detection algorithm using only dual-temporal images, the OA and kappa coefficients of the forest type classification results of Yunnan Province generated based on STM-CF increased by 10.06% and 0.1855 respectively. Therefore, compared with the existing sample migration method, the STM-CFS proposed in this paper can effectively filter out samples where forest type replacement has occurred between the historical year and the target year (such as pure forest to mixed forest), improve the quality of sample migration, and improve the classification accuracy of forest types in mountainous areas.

[0123] Table 3 Comparison of forest type classification accuracy of different sample migration methods

[0124]

[0125] from Figure 3-Figure 6 It can be seen that the effectiveness of the sample migration method proposed in this embodiment in remote sensing classification of mountain forest types is generally better than that of existing sample migration methods, but different sample migration methods have significant differences in the classification accuracy of each forest type; for example, the classification accuracy of coniferous forests and broad-leaved forests is significantly higher than that of mixed coniferous and broad-leaved bamboo forests. Compared with using all historical samples, the classification accuracy of the dual-temporal image change detection algorithm for coniferous forests, broad-leaved forests, and mixed coniferous and broad-leaved forests shows a significant upward trend, but the classification accuracy of bamboo forests decreases slightly; the PA of coniferous forests (such as Figure 3 As shown), UA (as Figure 4 as shown) and F1score (as shown Figure 5 The PA, UA and F1score of the broad-leaved forest increased by 0.95%, 15.69% and 0.0890 respectively; the PA and F1score parameters of the mixed coniferous and broad-leaved forest increased by 11.84% and 0.0018, while the UA decreased by 1.73%; the UA of the bamboo forest increased by 8.11%, while the PA and F1score decreased by 24.00% and 0.0037. Compared with the dual-temporal change detection method, the STM-CFS sample migration method proposed in this paper has significantly improved the classification accuracy of all forest types; for example, the PA, UA and F1score of coniferous forests increased by 12.05%, 6.29% and 0.0963, respectively; the PA, UA and F1score of broad-leaved forests increased by 7.87%, 5.34% and 0.0686, respectively; the PA, UA and F1score of mixed coniferous and broad-leaved forests increased by 29.40%, 37.37% and 0.3810, respectively; the PA, UA and F1score of bamboo forests increased by 12.10%, 63.19% and 0.6478, respectively. The above research results show that compared with directly applying all historical samples to the forest type classification of the target year, the use of dual-temporal image change detection methods to extract unchanged samples effectively increases the quality of training samples and improves the classification accuracy of coniferous forests, broad-leaved forests and mixed coniferous and broad-leaved forests; but due to the small number of bamboo forest samples (such as Figure 6 The bamboo forest samples detected by this method contain a certain proportion of false positive samples where the dominant tree species in the bamboo forest have changed, resulting in a significant increase in the omission error of the bamboo forest. The STM-CFS method effectively retains samples where there is no real forest type change, improves the quality of training samples, and improves the classification accuracy of each forest type.

[0126] It can be seen from the above technical solutions that the remote sensing classification method of forest types based on historical sample migration provided by the embodiment of the present invention can effectively use the forestry field survey samples of known historical years to infer the remote sensing classification map of forest types in the target year. Case studies in the monitoring area have shown that the method provided by the present invention can effectively improve the migration quality of historical forestry field survey samples, increase the number of effective samples in the target year, and improve the classification accuracy of mountain forest types under small sample conditions by machine learning algorithms. Compared with the existing remote sensing classification method of forest types based on historical sample migration, the remote sensing classification method of forest types of the present invention has higher quality of migration samples, higher consistency between the obtained samples and the forestry field survey results of the target year, higher remote sensing classification accuracy and precision, and higher effectiveness for the classification of forest types in a complex terrain area such as the monitoring area.

[0127] The above description is only a preferred embodiment of the present invention and an explanation of the technical principles used. It is not intended to limit the scope of the invention claimed for protection, but only represents the preferred embodiment of the present invention. Those skilled in the art should understand that the scope of the invention involved in the present invention is not limited to the technical solution formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the inventive concept. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

Claims

1. A remote sensing classification method for forest types based on historical sample migration, characterized in that: The method comprises the following steps: Step S1, determining the monitoring area, historical years and target years, obtaining historical sample points of forestry field surveys in historical years and target sample points of forestry field surveys in the target year within the monitoring area; Step S2, obtaining long-term multi-source remote sensing data of the historical sample points from the historical year to the target year; Step S3, using long-term multi-source remote sensing data to characterize the forest growth succession characteristics of historical sample points; Step S4, based on the forest growth succession characteristics, determine whether the forest type to which the historical sample point belongs has been replaced from the historical year to the target year; if the historical sample point has similar spectral similarity between the historical year and the target year, the temporal trajectory between the historical year and the target year, and the geographical environment similarity between the historical sample point and the sample point of the same type in the target year, then the forest type to which the historical sample point belongs has not been replaced; otherwise, it is determined that it has been replaced; Step S5, migrating the forestry field survey sample data of the historical sample points that have not been replaced to the target year, and deleting the historical sample points that have been replaced; Step S6, extracting classification features related to remote sensing classification of regional forest types using multi-source remote sensing data of the target year; Step S7, dividing the monitoring area into a plurality of floras according to the spatial distribution characteristics of the floras in the monitoring area, and selecting the optimal classification features for forest type classification of each flora from the relevant classification features using an association hierarchical clustering method; Step S8, based on the optimal classification features and the forestry field survey sample data of the historical sample points that have not been replaced and migrated to the target year, use the random forest algorithm to classify the forest type of each flora; Step S9, integrating the classification results of forest type classification of all floras to obtain a spatial distribution map of forest types in the entire monitoring area.

2. The forest type remote sensing classification method based on historical sample migration according to claim 1 is characterized in that: Step S1 is to obtain forestry field survey sample point data for historical years and target years in the monitoring area. First, the forestry field survey data for historical years and target years in the monitoring area are obtained, and then the historical sample point data and target sample point data maps are generated respectively using the surface-to-point method.

3. The forest type remote sensing classification method based on historical sample migration according to claim 1 is characterized in that: The long-term multi-source remote sensing data in step S2 include all available Landsat Collection 2 Tier 1 surface reflectance data images, Sentinel-1 / 2 data, SRTM-DEM data, climate data, and soil moisture data on the Google Earth Engine GEE platform corresponding to the period from historical years to target years.

4. The forest type remote sensing classification method based on historical sample migration according to claim 1 is characterized in that: The forest growth succession characteristics described in step S4 include the spectral similarity of the historical sample points in the historical year and the target year, the time series trajectory of the historical sample points from the historical year to the target year, and the environmental similarity of the historical sample points and the target sample points of the same type.

5. The forest type remote sensing classification method based on historical sample migration according to claim 4 is characterized in that: Step S4 determines whether the forest type of the historical sample point has changed from the historical year to the target year, using a threshold segmentation method.

6. The forest type remote sensing classification method based on historical sample migration according to claim 5 is characterized in that: The threshold segmentation formula is as follows: Moreover, Environmental similarity = max(S i1 , S i2 , …, S ij ) In formula (8), Forest type is the forest type label of the historical sample point; SAD and ED are the spectral similarity indicators that quantify the historical sample point between the historical year and the target year, σ1 and σ2 are the segmentation thresholds of SAD and ED respectively; disturbanceyear is the forest disturbance detection result of the historical sample point from the historical year to the target year, Historical year is the historical year; Environmental similarity is the environmental similarity between the historical sample point and the target sample point of the same type, σ3 is the segmentation threshold of environmental similarity, S i,j is the environmental similarity between all sample points of the same forest type in the historical sample point i and the target sample point j.

7. The forest type remote sensing classification method based on historical sample migration according to claim 6 is characterized in that: The value of σ1 is 0.95; the value of σ2 is 0.

1.

8. The forest type remote sensing classification method based on historical sample migration according to claim 1 is characterized in that: The migration model of forestry field survey sample data of historical sample points in step S5 is: In formula (9), the forest type historical sample point is the forest type of the historical sample point in each historical forestry field survey sample from the historical year to the target year. If the forest type of the historical sample point has not changed, it means that the forest type label of the sample point has not changed, and it is migrated to the target year for forest type remote sensing classification; otherwise, the type label information of the sample point has changed, and the sample point is deleted.

9. The forest type remote sensing classification method based on historical sample migration according to claim 7 is characterized in that: The relevant classification features extracted in step S6 include: spectrum, polarization, texture, phenology and environment.

10. The forest type remote sensing classification method based on historical sample migration according to any one of claims 1 to 9, characterized in that: The method further comprises: Step S10, evaluating mapping accuracy using forestry field survey sample data of the target year.

Images