CN116503681A - Crop identification feature optimization method and system for plain and hilly areas - Google Patents

Crop identification feature optimization method and system for plain and hilly areas Download PDF

Info

Publication number
CN116503681A
CN116503681A CN202310453932.0A CN202310453932A CN116503681A CN 116503681 A CN116503681 A CN 116503681A CN 202310453932 A CN202310453932 A CN 202310453932A CN 116503681 A CN116503681 A CN 116503681A
Authority
CN
China
Prior art keywords
landscape
crop
index
landscape mode
area
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310453932.0A
Other languages
Chinese (zh)
Inventor
张焕雪
李江明月
王琛
王昊平
陈宸
房冠儒
宋文瑶
张浩彧
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Normal University
Original Assignee
Shandong Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong Normal University filed Critical Shandong Normal University
Priority to CN202310453932.0A priority Critical patent/CN116503681A/en
Publication of CN116503681A publication Critical patent/CN116503681A/en
Pending legal-status Critical Current

Links

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
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/771—Feature selection, e.g. selecting representative features from a multi-dimensional feature space
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • 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
    • G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
    • G06V10/422—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation for representing the structure of the pattern or shape of an object therefor
    • 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
    • G06V10/52—Scale-space analysis, e.g. wavelet analysis
    • 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
    • G06V10/54—Extraction of image or video features relating to texture
    • 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
    • G06V10/58—Extraction of image or video features relating to hyperspectral data
    • 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
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
    • 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
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/188—Vegetation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
    • Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Artificial Intelligence (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Remote Sensing (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a crop identification characteristic optimization method and a system for plain and hilly areas, comprising the following steps: making a lookup table, and storing the optimal feature set of the drawing unit and the corresponding landscape mode label into the lookup table; acquiring a landscape mode of a farmland landscape area to be inquired, and outputting a drawing unit and an optimal feature set corresponding to the current landscape mode according to records in a lookup table; and (3) adopting the optimal feature set and the classifier of the corresponding drawing unit to finish the crop classification task of the landscape mode of the farmland landscape area to be inquired. The method is suitable for crop classification and cartography unit optimization in plain and hilly areas, and has high feasibility, robustness and classification result accuracy.

Description

Crop identification feature optimization method and system for plain and hilly areas
Technical Field
The invention belongs to the technical field of image processing in farmland landscape areas, and particularly relates to a crop identification characteristic optimization method and system for plain and hilly areas.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
The spatial distribution of crop types is critical for crop growth monitoring, yield estimation and disaster assessment. The remote sensing technology has proved to be a powerful data source for crop type drawing, and has the advantages of wide coverage, timely data acquisition, high updating speed, dynamic updating and the like. Recently, there is increasing effort on how to improve the accuracy and efficiency of crop classification from the aspects of mapping units, feature selection, classification algorithms, etc.
When the crop types of the investigation region are different, the optimal characteristics are also different, even for the same crop type, when the crop spatial distribution of the investigation region is different. Therefore, it is necessary to conduct systematic experiments to determine the best characteristics of the specific crop planting structure and spatial distribution.
The existing cartography units for crop classification, pixel-based and object-based methods employed, take into account spectral and spatial features extensively. First, pixel-based classification is the most widely used crop mapping method, and when using low resolution remote sensing images, there is typically mixed pixel and "salt-and-pepper" noise. In order to solve these problems, an object-based image analysis (OBIA) method has been developed and employed, especially when a high resolution image is used; secondly, classification accuracy is improved by combining spectral and spatial features, especially for crop types with similar spectral features.
However, both of the above solutions have drawbacks:
in a first aspect, while both pixel-based mapping and object-based classification achieve similar crop type recognition accuracy, pixel-based classification uses fewer features and requires less computation time. I.e. a trade-off between estimation accuracy and data processing complexity needs to be considered when selecting the mapping unit;
in a second aspect, the optimal characteristics of the same crop will vary with the area of investigation. Therefore, systematic studies are required to determine the optimal spectral and spatial characteristics of a particular region;
in a third aspect, the landscape architecture is described primarily from the perspective of the spatial distribution of crops (i.e., crop type or field size) in the area of investigation. Few tools or metrics are considered for comprehensive quantitative assessment of agricultural landscapes. While these studies demonstrate that landscape indices are powerful tools for spatial topography analysis, which can be used to evaluate agricultural landscapes, the use of all indices is inferior to the selection of a set of indices when redundancy issues are considered. Therefore, it is necessary to determine the most comprehensive, representative and non-redundant indicators to characterize crop heterogeneity.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a crop identification characteristic optimization method and a crop identification characteristic optimization system for plain and hilly areas.
To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:
the first aspect of the invention provides a crop identification feature optimization method for plain and hilly areas;
making a lookup table, and storing the optimal feature set of the drawing unit and the corresponding landscape mode label into the lookup table;
the step of obtaining the optimal feature set of the drawing unit comprises the following steps: acquiring remote sensing images of P different landscape mode labels of a known farmland landscape area;
preprocessing the remote sensing image to generate a sample set; extracting L image features on a pixel-based scale and an object-based scale for the remote sensing image of each landscape mode in the sample set;
based on the L image features, M feature classification schemes of the known landscape mode labels are formulated;
classifying M feature classification schemes of the known landscape mode labels on the scale based on pixels and the scale based on objects respectively to obtain an optimal feature classification scheme; taking the characteristics corresponding to the optimal characteristic classification scheme as an optimal characteristic set;
acquiring a landscape mode of a farmland landscape area to be inquired, and outputting a drawing unit and an optimal feature set corresponding to the current landscape mode according to records in a lookup table;
and (3) adopting the optimal feature set and the classifier of the corresponding drawing unit to finish the crop classification task of the landscape mode of the farmland landscape area to be inquired.
In a second aspect the invention provides a crop identification feature preference system for plain and hilly areas comprising:
a look-up table formulation module configured to: making a lookup table, and storing the optimal feature set of the drawing unit and the corresponding landscape mode label into the lookup table;
the step of obtaining the optimal feature set of the drawing unit comprises the following steps: acquiring remote sensing images of P different landscape mode labels of a known farmland landscape area;
preprocessing the remote sensing image to generate a sample set; extracting L image features on a pixel-based scale and an object-based scale for the remote sensing image of each landscape mode in the sample set;
based on the L image features, M feature classification schemes of the known landscape mode labels are formulated;
classifying M feature classification schemes of the known landscape mode labels on the scale based on pixels and the scale based on objects respectively to obtain an optimal feature classification scheme; taking the characteristics corresponding to the optimal characteristic classification scheme as an optimal characteristic set;
a lookup module configured to: acquiring a landscape mode of a farmland landscape area to be inquired, and outputting a drawing unit and an optimal feature set corresponding to the current landscape mode according to records in a lookup table;
a crop classification module configured to: and (3) adopting the optimal feature set and the classifier of the corresponding drawing unit to finish the crop classification task of the landscape mode of the farmland landscape area to be inquired.
A third aspect of the invention provides a computer readable storage medium having stored thereon a program which when executed by a processor performs the steps of a method of crop identification feature preference for plain and hilly areas according to the first aspect of the invention.
A fourth aspect of the invention provides an electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in a crop identification feature optimization method for plain and hilly areas according to the first aspect of the invention when the program is executed.
The one or more of the above technical solutions have the following beneficial effects:
(1) The invention provides a method for evaluating the heterogeneity of crops by developing a unified framework, and analyzing the influence of the heterogeneity on optimization characteristics and classification units. The characteristic preferred method of the invention comprises the following steps: optimizing the landscape index in the research by utilizing correlation analysis and factor analysis; dividing a landscape area by using a K-Means cluster analysis method, and synthesizing factor analysis results to determine different landscape mode labels; screening based on a random forest algorithm and determining an optimal feature subset according to importance evaluation; constructing a lookup table to search the best classification feature and classification unit in each landscape group; and the vegetation index and texture characteristics are increased, and the precision loss caused by configuration heterogeneity is properly compensated. The method is suitable for plain and hilly area crop classification and cartography unit optimization, has high feasibility, robustness and classification result accuracy, and can provide guidance and advice for improving effective large-scale crop mapping.
(2) In the invention, the spatial distribution of crops is considered, and systematic experiments are carried out to determine the drawing unit of the landscape of the specific crops; the landscape index can be optimized according to the correlation analysis and the factor analysis; the K-Means cluster analysis method is used for dividing the landscape area, and the result of factor analysis is synthesized, so that different landscape mode labels can be determined; in order to evaluate the importance of the classification features and the influence of the drawing units on the classification precision, a lookup table is constructed to search the optimal classification features and the drawing units in each landscape group, and if the landscape mode is known, the optimal classification features and the optimal drawing units can be selected through the lookup table, so that the highest classification precision is ensured to be achieved in the shortest time; the vegetation index and texture characteristics are added, so that the precision loss caused by configuration heterogeneity can be properly compensated; the feature subset is screened by adopting a random forest algorithm, so that the performance of high landscape heterogeneity can be improved; when the optimal requirements of the characteristics and the drawing units in the heterogeneous landscape on the classification performance of crops are determined, a set of system and comprehensive test are established. The method is suitable for crop classification and cartography unit optimization in plain and hilly areas, and has high feasibility, robustness and classification result accuracy.
Additional aspects of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a flow chart of a preferred method of identifying crop features in plain and hilly areas according to a first embodiment;
FIG. 2 (a) is a data diagram of the observation site in the first embodiment;
fig. 2 (b), 2 (c), 2 (d) are remote sensing images at different positions in fig. 2 (a), respectively;
FIG. 3 is an illustration of the importance of each feature of each view heterogeneous region on a pixel-based and object-based scale in a first embodiment;
fig. 4 (a) and 4 (b) are overall accuracy versus graphs of five classification schemes based on pixel scale and on object scale in the first embodiment, respectively;
FIGS. 5 (a), 5 (b), and 5 (c) are trends relating to Factor-1 and various features, respectively;
FIGS. 5 (e), 5 (f), and 5 (g) are trends relating to Factor-2 and various features, respectively;
fig. 5 (h), 5 (i), 5 (j) are the relative trends between the various features, respectively;
FIGS. 6 (a) -6 (e) and 6 (f) -6 (j) are correlations between Factor-1 and Factor-2, respectively, and the results of the accuracy assessment of each classification scheme on a pixel scale;
FIGS. 6 (k) -6 (o) and 6 (p) -6 (t) are correlations between Factor-1 and Factor-2, respectively, and the accuracy assessment results for each classification scheme on the object scale;
fig. 7 (a) to 7 (d) show differences between component heterogeneity and arrangement heterogeneity, respectively.
Detailed Description
Example 1
As shown in fig. 1, the present embodiment discloses a crop identification feature optimization method for plain and hilly areas, which includes:
step S1: making a lookup table, and storing the optimal feature set and the landscape mode label of the corresponding drawing unit into the lookup table;
step S2: acquiring a landscape mode of a farmland landscape area to be inquired, and outputting a drawing unit and an optimal feature set corresponding to the current landscape mode according to records in a lookup table;
step S3: and (3) adopting the optimal feature set and the classifier of the corresponding drawing unit to finish the crop classification task of the landscape mode of the farmland landscape area to be inquired.
In step S1, the optimal feature set acquisition step of the corresponding drawing unit includes:
s101: acquiring remote sensing images of P different landscape mode labels of a known farmland landscape area; the value of P is 4;
s102: preprocessing a remote sensing image to generate a sample set; extracting L image features on a pixel-based scale and an object-based scale for the remote sensing image of each landscape mode in the sample set; l takes a value of 48;
s103: based on L image features, M feature classification schemes of the known landscape mode labels are formulated; m has a value of 5;
s104: classifying M feature classification schemes of the known landscape mode labels on the scale based on pixels and the scale based on objects respectively to obtain an optimal feature classification scheme; and taking the characteristics corresponding to the optimal characteristic classification scheme as an optimal characteristic set.
In S101, acquiring remote sensing images of P different landscape mode tags in a known farmland landscape area, including:
s101-1: selecting a landscape index derived from the Sentinel-2 data;
wherein the landscape indices substantially reflect the characteristics of the crop distribution in the investigation region and these indices are most important for characterizing agricultural landscapes and reflecting the most important characteristics of the crop distribution. Therefore, the landscape index is selected by considering the landscape space structure, the space characteristics, the landscape diversity and the physical meaning of the index.
S101-2: optimizing a landscape index;
correlation Analysis (CA) and Factor Analysis (FA) were used to optimize the landscape index in the study.
The Spearman correlation coefficient was used as an index for CA and the significance of the correlation coefficient was checked using a two-tailed t-test (significance levels 0.05 and 0.01). The rank correlation coefficient is calculated as follows:
wherein n is two variationsThe rank logarithm of the quantity x, y, i.e. the sample content. d, d i Is the difference of the same pair of ranks (i=1, 2,..n).
In FA, several closely related variables are classified into the same category, each category becoming a factor. The FA ends by solving the initial factor loading matrix, constructing the factor model, and performing a rotation transformation on the factor loading matrix. This factor rotation makes the contribution of each factor to each principal component more clear and its significance easier to analyze.
The mathematical form is:
X＝A×F+E
F＝S×X+E
wherein X is an original variable vector, A is a common factor load matrix, F is a common factor vector, E is a common factor which indicates that the influence of factors on data variance is negligible, and S is a factor score matrix. Factor scores for F were also estimated to specify the magnitude of the contribution of the variable to the common factor.
In a specific implementation, first, a principal component with a feature root greater than 1, namely a common factor class, is selected. And then obtaining A and S through the FA, and determining the contribution of each type of common factors according to the relatively independent landscape indexes. The importance of each class of common factors is determined by their contribution to the relatively independent landscape index, which is filtered to obtain an optimal index describing the agricultural landscape.
S101-3: evaluating crop heterogeneity based on the optimized landscape index;
evaluating crop heterogeneity based on the optimized landscape index; the method specifically comprises the following steps:
first, 8,362 plots were divided into landscape areas with similar objects using the K-Means cluster analysis method.
Illustratively, to better describe and make more representative crop heterogeneity, the full province is divided into a grid of 5km by 5 km. Only grids with more than 30% of the cultivated area were reserved for investigation (distributed in plain and hills). FIG. 2 (a) is data of an observation site applied to an embodiment of the present invention; fig. 2 (b), 2 (c) and 2 (d) are enlarged views of data observed at different positions in fig. 2 (a), respectively.
S101-4: determining different landscape mode labels based on crop heterogeneity assessment indexes (component heterogeneity Factor-1 and configuration heterogeneity Factor-2);
wherein, different view mode labels include:
the average plaque area is larger and uniform, the crop variety is less, the landscape plaque distribution is discrete, and the geometric shape is simple in 1098 grid number areas: A1B1;
the average plaque area is larger and uniform, the crop variety is less, the landscape plaque distribution is concentrated, and 2516 grid number areas with complex geometric shapes are formed: A1B2;
the average plaque area is smaller and is staggered, the crop variety is more, the landscape plaque distribution is discrete, and the geometric shape is simple in 2018 grid number areas: A2B1;
the average plaque area is smaller and is staggered, the crop variety is more, the landscape plaque distribution is concentrated, and 2730 grid number areas with complex geometric shapes are formed: A2B2;
the invention collects 15 Sentinel-2 image blocks with cloud content lower than 10% in 2019 crop growing season, and uses 9 wave bands of each image as data input. Using Sen2Cor v2.9, atmospheric Top (TOA) data was first converted to atmospheric Bottom (BOC) reflectivity, then bilinear interpolation was used. Finally, the image covering the investigation region is mosaic and cropped based on the grid boundaries. Ground truth data was collected from 7 months to 9 months in 2019. During the investigation, detailed information of representative crop types (rice, corn and soybean) was recorded, and geographical position location accuracy of less than 5m was recorded for each sample using hand-held GPS. The number of corn, rice and soybean samples throughout the study area were 2,088, 1,818 and 1,223, respectively. And less than 100 samples were used to supplement the grid by manual visual interpretation.
For the number of landscape areas, the area with the closest grid number was selected by five repeated experiments. Crop heterogeneity within each region is described as component heterogeneity and configuration heterogeneity. The former refers to an increase in the number of crop species, and the latter refers to an increase in the degree of spatial dispersion of the crop. Due to the large number of grids, 1% of grids in each area are randomly selected as the study object of the subsequent experiment.
In S102, preprocessing the remote sensing image to generate a sample set; the method specifically comprises the following steps:
the reflectivity of the remote sensing image is converted from the top of the atmosphere to the bottom of the atmosphere;
sampling the converted image in a bilinear interpolation mode;
and cutting the sampled data to obtain an image of the research area.
Illustratively, acquiring the second remote sensing image data of the sentinel covering the area, and preprocessing the image data, including: atmospheric correction, format conversion and image clipping according to farmland maps. And generating a sample set according to the field survey data and combining google earth images.
Acquiring sentinel second-number remote sensing image data covering the area, preprocessing the image data, and comprising the following steps:
atmospheric Top (TOA) data were first converted to atmospheric Bottom (BOC) reflectivity using Sen2Cor v2.9, followed by bilinear interpolation. Finally, the image covering the investigation region is mosaic and cropped based on the grid boundaries.
Generating a sample set in combination with google earth images, comprising: and downloading a map of a corresponding level by using a Google Earth map downloading tool, and selecting the ROI on the Google Earth image in the tif format by using ENVI5.3 image processing software and saving the ROI to a shape file for subsequent use.
Further, extracting L image features on pixel-based and object-based scales for the remote sensing images of each landscape mode in the sample set; the method specifically comprises the following steps:
and a multi-resolution fractal network evolution method is adopted to obtain spectrum and texture characteristics of the object scale.
A series of visual interpretations of the image were compared by trial and error.
The segmentation is performed according to the spectral bands of all images, with dimensions, shape parameters and compactness parameters of 50, 0.3 and 0.5, respectively.
Illustratively, the present study primarily considers three classes of features. A total of 23 vegetation indices and eight texture features were selected from the gray level co-occurrence matrix (GLCM). Texture was calculated using a sliding window of 9 x 9 pixels, and RE and NIR bands were chosen because of their high crop identification potential. Of the RE bands, only the B6 band is selected because it is superior to the B5 band in terms of crop identification. On this basis, the distance parameter of one pixel is selected to compare neighboring pixels, and 64-level gray scale quantization is derived. Texture was calculated in four directions (0 °, 45 °, 90 ° and 135 °), and rotational variance was achieved using the average of the four directions. A total of 48 features (including 9 spectral bands, 23 vegetation indices, and 16 texture features) were calculated on a pixel-based and object-based scale, respectively. For object-based dimensions, an average value of each feature of each object is used.
The 48 features refer to 9 spectral bands, 23 vegetation indexes and 16 texture features.
Wherein 9 spectral bands are respectively Band 2, band 3, band 4, band 5, band 6, band 7, band 8, band 11 and Band 12, and are correspondingly denoted as B2, B3, B4, B5, B6, B7, B8, B11 and B12, wherein B refers to Band.
Wherein the 23 vegetation indices comprise a spectral vegetation index without RE reflectivity and a spectral vegetation index with RE reflectivity;
spectral vegetation indices without RE reflectivity include: green Index (GI), normalized vegetation index (NDVI), green vegetation index (VIgreen), reformed vegetation index (RDVI), soil-adjusted vegetation index (SAVI), corrected simple ratio (MSR), simple ratio index (RVI), blue-green ratio index (BGI), red-green ratio index (RGI), differential Vegetation Index (DVI), green NDVI (GNDVI), improved soil-adjusted vegetation index (MSAVI);
spectral vegetation indices with RE reflectivity include: b5 RedEdge-near infrared NDVI (B5 rednvi), B6 RedEdge-near infrared NDVI (B6 rednvi), B5 converted CARI (B5 TCARI), B6 converted CARI (B6 TCARI), B5 triangular vegetation index (B5 TVI), B6 triangular vegetation index (B6 TVI), red edge NDVI (RNDVI), green-red edge NDVI (GRNDVI), modified Chlorophyll Absorption Reflectance Index (MCARI), red edge ratio 1 (RRI 1), red edge ratio 2 (RRI 2).
Wherein, the texture feature includes: b5 variance (Var), B5 homogeneity (Hom), B5 contrast (Con), B5 dissimilarity (Dis), B5 entropy (Ent), B5 Angular Second Moment (ASM), B5 correlation (Cor), B6 variance (Var), B6 homogeneity (Hom), B6 contrast (Con), B6 dissimilarity (Dis), B6 entropy (Ent), B6 Angular Second Moment (ASM), B6 correlation (Cor).
Table 1 initial feature set in the study
In S103, based on the L image features, M feature classification schemes of the known landscape mode tag are formulated, specifically including:
scheme a: the S scheme is short, and the reflectivity of 9 spectral bands (B2, B3, B4, B5, B6, B7, B8, B11 and B12) is shown in the specification;
scheme B: the S+V scheme is a combination of 32 spectral features (9 spectral bands+23 vegetation indexes);
scheme C: the S+G scheme is a combination of 25 features (9 spectral bands+16 texture features);
scheme D: the ALL scheme is a combination of ALL 48 features (9 spectral bands+23 vegetation indexes+16 texture features);
scheme E: the FS scheme for short, is the best subset of 48 features.
Five classification schemes were designed to perform crop extraction and further evaluate the performance of different feature combinations.
Specifically, scheme a uses reflectivity of 9 spectral bands to map the crop planting area, scheme B uses only 32 spectral features, scheme C uses 9 spectral bands in combination with 16 texture features, scheme D uses all 48 features, and scheme E proceeds based on the results of feature selection.
Further, scheme E, determination of the optimal subset of 48 features; the method specifically comprises the following steps:
all 48 image features are simultaneously input into a random forest algorithm, and weights of different features are output;
sorting the features according to the order of the weights from large to small;
the top-ranked features are selected and the combination is referred to as the best feature subset.
The weights and rankings of the different feature variables are evaluated using the random forest software package in Matlab2018 to select the best feature subset.
G＝Uniform({g t })
The optimal dimensions of candidate features ranked by importance were studied by Sequence Forward Selection (SFS).
Because of the different sensitivity to specific features, not all features are equally effective in identifying crops. Numerous studies have shown that random forest algorithms are attractive not only in classification, but also in feature selection. The present invention uses the random forest software package in Matlab2018 to evaluate the weights and ordering of the different feature variables to select the best feature subset. The inherent randomness of the random forest method is considered to lead to uncertainty in the results of each feature.
The optimal dimensions of candidate features ranked by importance were studied by Sequential Forward Selection (SFS). The process operates by adding features one after the other until the precision seed increases significantly with the addition of new features, indicating that relatively high precision and low computational costs can be achieved by reducing the amount of data.
The random forest is used for feature selection, and feature selection is carried out according to the obtained sequencing result of the feature set by adopting a method for carrying out importance measurement on feature vectors. The specific construction steps are as follows:
the number of original samples is N, and the characteristic dimension is m.
Input: a set of training samples: { (x) 1 ,y 1 ),...,(x N ,y N )}；
Step 1: generating a self-service sample set D with the size of N t 。
Step 2: in self-help sample set D t G is obtained through a classification regression tree algorithm t 。
Step 3: sampling the columns (features) of each tree, and randomly selectingAnd is characterized in that n try And selecting a feature with highest variable importance to perform node SPLIT.
Step 4: judging whether the number t of the current tree satisfies t is less than or equal to n tree If yes, repeating the steps 1-3; if not, the cycle is stopped.
And (3) outputting: g=form ({ G) t })。
And sequentially starting the feature subsets from the empty set, and selecting one feature at a time to add the feature subsets so as to optimize the feature function. The essence is that a feature addition is selected each time so that the value of the evaluation function is optimized. And obtaining an iterative feature evaluation result, which is used for solving the problems of randomness in determining the size of the feature subset and instability of the result in the existing algorithm. That is, according to the results of the feature importance descending order, the next feature is added from the feature located at the first position, then, for each feature vector combination, the classification error probability is calculated, until all the features are used to obtain the classification error probability of the classification model, and the combination with the minimum error probability is selected as the final feature selection result. (same)
Wherein the RF classifier requires two main parameters to generate the predictive model: the number of classification trees required (n tree ) And the number of prediction variables (m try ). In our study, several RF models were constructed using selected features, n for each model tree Set to 1000, m try Equal to the square root of the number of input variables.
The accuracy of general crop classification can be assessed using a confusion matrix.
The Overall Accuracy (OA) is used to evaluate the crop classification accuracy of the classification scheme and the mapping unit. For comparability, the same training and validation samples were used for different classification scales, and the same classifier parameter settings were used for both data scales. All of this (including the RF classifier and confusion matrix) is implemented using the "ENMAP" plugin component under version 5.3 of ENVI. Fig. 4 (a) and 4 (b) are overall accuracy versus graphs of five classification schemes based on pixel scale and on object scale in the first embodiment, respectively;
in general, landscape heterogeneity includes constituent heterogeneity and configuration heterogeneity. In the present invention, five indices, namely, SPLIT, frac_ AM, SHEI, AREA _mn and enn_mn, represent non-uniformity of composition and configuration by excluding landscape indices having high correlation while retaining at least one landscape index of each type. SPLIT refers to the complexity of the spatial structure of the type of land cover in the corresponding landscape, frac_am refers to the dimension of the division, she refers to the fragrance uniformity index, area_mn refers to the integrated measure of the number and AREA of the landscape type, and en_mn refers to the european nearest distance. (different indexes)
And obtaining the complexity of the space structure of the land cover type in the regional landscape by adopting the SPLIT index. The calculation formula is as follows: (open)
Wherein A is the total landscape area, a ij The area of the land crop i is j, and k is the plaque number in the crop i. When the landscape is composed of a single patch, split=1, the smaller the patch contained in the landscape is, the larger the division is, and the division value reaches the maximum value when the landscape is subdivided to the maximum. Lower SPLIT means less conformational heterogeneity and vice versa.
Meanwhile, the complexity of plaque shapes in the regional landscape is obtained by using the FRAC_AM index so as to quantitatively describe the landscape pattern. The calculation formula is as follows:
D＝2log(P/4)/log(A)
wherein: d represents the dimension of the division, P is the plaque perimeter, and A is the plaque area. The theoretical range of the fractal dimension values in landscape ecology is that the fractal dimension of 1.0-2.0 approaches 1, and the plaque shape is more regular. On the other hand, the score approaches 1, indicating that the plaque geometry approaches a simple square or circle. 2.0 represents the most complex plaque type with the same area circumference, which is typically worth the possible upper limit of 1.5. The higher the dimension of the partition, the more complex the geometry of the landscape.
Meanwhile, the SHEI index is also adopted to obtain the uniformity degree of distribution of different ecological systems in the regional landscape. The calculation formula is as follows:
SHEI＝(H/H max )
H max ＝og(m)
wherein: the SHEI is the relative uniformity index (%); h is Shannon-Weaver diversity index, H max Is the largest possible uniformity of the landscape, and m is the total number of landscape types. Uniformity of distribution of each plaque in the landscape in area, the more uniform the plaque area distribution, the more the uniformity tends to be 1.
Meanwhile, the AREA_MN index is also used for obtaining the comprehensive measure of the quantity and the AREA of the regional landscape types. The calculation formula is as follows:
wherein: area_mn is equal to the total AREA of a certain plaque type divided by the number of plaques of that type at the plaque level; equal to the total area of the landscape divided by the total number of patches of each type at the landscape level. a, a ij Is the area of plaque ij, n i Is the number of all relevant patches in the landscape type, and N is the total number of patches in the landscape as a whole. A lower AREA _ MN means a stronger fragmentation and vice versa.
Meanwhile, the ENN_MN index is also adopted to obtain the spatial pattern of the regional measurement landscape. The formula is as follows:
the ENN_MN value is large, the distance between the plaques of the same type is far, and the distribution is discrete; otherwise, the close distance between the plaques of the same type is indicated to be in agglomeration distribution.
In each area, the non-crop pixels are masked by a farmland map generated from GLC10 (Global Land Cover classes:10 meter resolution global land cover map). To compare their heterogeneity, all regions were of the same size.
The importance of each feature of each landscape heterostructure based on pixel and object based scale is derived using SFS methods (as shown in fig. 3). To construct a look-up table (e.g., table 2) to quickly search for optimal features at pixel-based and object-based scales, respectively, for different heterogeneities, the study area is divided into 4 landscape groups; each region location of the landscape group is determined based on K-Means cluster analysis, and the number of landscape groups is determined based on five repeated experiments to select the region with the closest grid number.
Table 2 look-up table
Determining optimal features and mapping units for different heterogeneities, comprising: according to the experimental result analysis of the examples, the component heterogeneity and the configuration heterogeneity of each group were judged. And respectively calculating the average value of the five optimal landscape indexes, comparing the result analysis, and determining the optimal characteristics and the drawing unit.
The five best landscape indexes are calculated according to the drawing result with highest classification precision of each region. Thus, it can be considered that the real landscape heterogeneity of all the areas is obtained. In each area, non-crop pixels are masked by the farmland map generated from GLC 10. To compare their heterogeneity, all regions were of the same size.
In order to construct a lookup table to quickly search for the best features and drawing units under different heterogeneities, 8362 plots are divided into 4 landscape modes according to the cluster analysis of the K-Mean method and the physical significance of five selected landscape indexes on the premise of FA analysis.
The average plaque area is larger and uniform, the crop variety is less, the landscape plaque distribution is discrete, and the geometric shape is simple in 1098 grid number areas: A1B1;
the average plaque area is larger and uniform, the crop variety is less, the landscape plaque distribution is concentrated, and 2516 grid number areas with complex geometric shapes are formed: A1B2;
the average plaque area is smaller and is staggered, the crop variety is more, the landscape plaque distribution is discrete, and the geometric shape is simple in 2018 grid number areas: A2B1;
the average plaque area is smaller and is staggered, the crop variety is more, the landscape plaque distribution is concentrated, and 2730 grid number areas with complex geometric shapes are formed: A2B2;
each region location of the landscape group is determined based on K-Means cluster analysis, and the number of landscape groups is determined based on five repeated experiments to select the region with the closest grid number.
FIGS. 5 (a), 5 (b), and 5 (c) are trends relating to Factor-1 and various features, respectively; FIGS. 5 (e), 5 (f), and 5 (g) are trends relating to Factor-2 and various features, respectively; fig. 5 (h), 5 (i), and 5 (j) are the relative trends between the various features, respectively. FIGS. 6 (a) -6 (e) and 6 (f) -6 (j) are correlations between Factor-1 and Factor-2, respectively, and the results of the accuracy assessment of each classification scheme on a pixel scale; fig. 6 (k) to 6 (o) and fig. 6 (p) to 6 (t) are correlations between Factor-1 and Factor-2, respectively, and the accuracy evaluation results of each classification scheme on the object scale.
The best features and mapping units for the different heterogeneities are determined. According to the experimental result analysis of the example, the A1B1 group is characterized by the region that the composition heterogeneity and the configuration heterogeneity are low, the A1B2 group is characterized by the region that the composition heterogeneity and the configuration heterogeneity are high, the A2B1 group is characterized by the region that the composition heterogeneity and the configuration heterogeneity are low, and the A2B2 group is characterized by the region that the composition heterogeneity and the configuration heterogeneity are high. And respectively calculating the average value of the five optimal landscape indexes, comparing the result analysis, and determining the optimal characteristics and the drawing unit. Fig. 7 (a) to 7 (d) show differences between component heterogeneity and arrangement heterogeneity, respectively.
Example two
The present embodiment discloses a crop identification feature preference system for plain and hilly areas, comprising:
a look-up table formulation module configured to: making a lookup table, and storing the optimal feature set of the drawing unit and the corresponding landscape mode label into the lookup table;
the step of obtaining the optimal feature set of the drawing unit comprises the following steps: acquiring remote sensing images of P different landscape mode labels of a known farmland landscape area;
preprocessing the remote sensing image to generate a sample set; extracting L image features on a pixel-based scale and an object-based scale for the remote sensing image of each landscape mode in the sample set;
based on the L image features, M feature classification schemes of the known landscape mode labels are formulated;
classifying M feature classification schemes of the known landscape mode labels on the scale based on pixels and the scale based on objects respectively to obtain an optimal feature classification scheme; taking the characteristics corresponding to the optimal characteristic classification scheme as an optimal characteristic set;
a lookup module configured to: acquiring a landscape mode of a farmland landscape area to be inquired, and outputting a drawing unit and an optimal feature set corresponding to the current landscape mode according to records in a lookup table;
a crop classification module configured to: and (3) adopting the optimal feature set and the classifier of the corresponding drawing unit to finish the crop classification task of the landscape mode of the farmland landscape area to be inquired.
Example III
An object of the present embodiment is to provide a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of a crop identification feature optimization method for plain and hilly areas as set forth in embodiment 1 of the present disclosure.
Example IV
An object of the present embodiment is to provide an electronic apparatus.
An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in a crop identification feature optimization method for plain and hilly areas as set forth in embodiment 1 of the present disclosure when the program is executed.
The steps involved in the devices of the second, third and fourth embodiments correspond to those of the first embodiment of the method, and the detailed description of the embodiments can be found in the related description section of the first embodiment. The term "computer-readable storage medium" should be taken to include a single medium or multiple media including one or more sets of instructions; it should also be understood to include any medium capable of storing, encoding or carrying a set of instructions for execution by a processor and that cause the processor to perform any one of the methods of the present invention.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented by general-purpose computer means, alternatively they may be implemented by program code executable by computing means, whereby they may be stored in storage means for execution by computing means, or they may be made into individual integrated circuit modules separately, or a plurality of modules or steps in them may be made into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.

Claims (10)

1. A method of optimizing crop identification characteristics in plain and hilly areas, comprising:
making a lookup table, and storing the optimal feature set of the drawing unit and the corresponding landscape mode label into the lookup table;
the step of obtaining the optimal feature set of the drawing unit comprises the following steps: acquiring remote sensing images of P different landscape mode labels of a known farmland landscape area;
preprocessing the remote sensing image to generate a sample set; extracting L image features on a pixel-based scale and an object-based scale for the remote sensing image of each landscape mode in the sample set;
based on the L image features, M feature classification schemes of the known landscape mode labels are formulated;
classifying M feature classification schemes of the known landscape mode labels on the scale based on pixels and the scale based on objects respectively to obtain an optimal feature classification scheme; taking the characteristics corresponding to the optimal characteristic classification scheme as an optimal characteristic set;
acquiring a landscape mode of a farmland landscape area to be inquired, and outputting a drawing unit and an optimal feature set corresponding to the current landscape mode according to records in a lookup table;
and (3) adopting the optimal feature set and the classifier of the corresponding drawing unit to finish the crop classification task of the landscape mode of the farmland landscape area to be inquired.
2. A method of optimizing crop identification characteristics in plain and hilly areas according to claim 1, wherein said obtaining remote sensing images of P different landscape mode tags for known farmland landscape areas comprises:
the correlation analysis CA and the factor analysis FA are utilized to optimize the landscape index;
evaluating crop heterogeneity based on the optimized landscape index;
determining different landscape mode labels based on the evaluation index of the crop heterogeneity;
the different landscape mode tags include:
the average plaque area is larger and uniform, the crop variety is less, the landscape plaque distribution is discrete, and the geometric shape is simple in 1098 grid number areas: A1B1;
the average plaque area is larger and uniform, the crop variety is less, the landscape plaque distribution is concentrated, and 2516 grid number areas with complex geometric shapes are formed: A1B2;
the average plaque area is smaller and is staggered, the crop variety is more, the landscape plaque distribution is discrete, and the geometric shape is simple in 2018 grid number areas: A2B1;
the average plaque area is smaller and is staggered, the crop variety is more, the landscape plaque distribution is concentrated, and 2730 grid number areas with complex geometric shapes are formed: A2B2.
3. A method of optimizing crop identification characteristics in plain and hilly areas according to claim 2, wherein said optimizing the landscape index using correlation analysis CA and factor analysis FA comprises:
calculating a grade correlation coefficient by using correlation analysis CA, and selecting a main component with a grade correlation coefficient characteristic root larger than 1, namely a common factor class;
the factors are analyzed to obtain A and S, and the contribution of each type of common factors is determined according to the relatively independent landscape indexes; the importance of each class of common factors is determined by their contribution to the relatively independent landscape index, which is filtered to obtain an optimal landscape index describing the agricultural landscape.
4. A method of crop identification feature preference for plain and hilly areas as claimed in claim 2, characterized in that the assessing crop heterogeneity based on optimized landscape index comprises:
using a K-Means cluster analysis method to divide 8,362 plots into landscape areas with similar objects;
dividing the whole province into grids of 5km multiplied by 5km, and only reserving grids with the cultivated area of more than 30%;
the constituent heterogeneity Factor-1 and the configuration heterogeneity Factor-2 within each grid region are calculated separately.
5. A method of crop identification preference for plain and hilly areas as claimed in claim 1, wherein the L image features include 9 spectral bands, 23 vegetation indices and 16 texture features.
6. A method for optimizing crop identification features in plains and hilly areas according to claim 5, wherein the M feature classification schemes of the known landscape mode labels are formulated based on the L image features; comprising the following steps:
scheme a: reflectivity of 9 spectral bands;
scheme B: a combination of 9 spectral bands +23 vegetation indices;
scheme C: a combination of 9 spectral bands +16 texture features;
scheme D: a combination of all 48 features;
scheme E: is the best subset of 48 features.
7. A plains and hills area crop identification feature optimization method according to claim 5, characterized in that the 23 vegetation indexes comprise a spectral vegetation index without RE reflectivity and a spectral vegetation index with RE reflectivity;
spectral vegetation indices without RE reflectivity include: green index, normalized vegetation index, green vegetation index, reformed vegetation index, soil adjusted vegetation index, correction simple ratio, simple ratio index, blue-green ratio index, red-green ratio index, differential vegetation index, green normalized vegetation index, and improved soil adjusted vegetation index;
spectral vegetation indices with RE reflectivity include: band 5 near infrared normalized vegetation index, band 6 near infrared normalized vegetation index, band 5 converted chlorophyll absorption reflection index, band 6 converted chlorophyll absorption reflection index, band 5 triangular vegetation index, band 6 triangular vegetation index, red side normalized vegetation index, green-red side normalized vegetation index, modified chlorophyll absorption reflection index, red side ratio 1 and red side ratio 2.
8. A crop identification preference system for plain and hilly areas, characterized by: comprising the following steps:
a look-up table formulation module configured to: making a lookup table, and storing the optimal feature set of the drawing unit and the corresponding landscape mode label into the lookup table;
the step of obtaining the optimal feature set of the drawing unit comprises the following steps: acquiring remote sensing images of P different landscape mode labels of a known farmland landscape area;
preprocessing the remote sensing image to generate a sample set; extracting L image features on a pixel-based scale and an object-based scale for the remote sensing image of each landscape mode in the sample set;
based on the L image features, M feature classification schemes of the known landscape mode labels are formulated;
classifying M feature classification schemes of the known landscape mode labels on the scale based on pixels and the scale based on objects respectively to obtain an optimal feature classification scheme; taking the characteristics corresponding to the optimal characteristic classification scheme as an optimal characteristic set;
a lookup module configured to: acquiring a landscape mode of a farmland landscape area to be inquired, and outputting a drawing unit and an optimal feature set corresponding to the current landscape mode according to records in a lookup table;
a crop classification module configured to: and (3) adopting the optimal feature set and the classifier of the corresponding drawing unit to finish the crop classification task of the landscape mode of the farmland landscape area to be inquired.
9. A computer readable storage medium having stored thereon a program, which when executed by a processor, implements the steps of a method for identifying crop characteristics preferred for plain and hilly areas according to any of the claims 1-7.
10. Electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, characterized in that the processor implements the steps of a method for optimizing crop identification characteristics in plain and hilly areas according to any of the claims 1-7 when said program is executed.
CN202310453932.0A 2023-04-20 2023-04-20 Crop identification feature optimization method and system for plain and hilly areas Pending CN116503681A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310453932.0A CN116503681A (en) 2023-04-20 2023-04-20 Crop identification feature optimization method and system for plain and hilly areas

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310453932.0A CN116503681A (en) 2023-04-20 2023-04-20 Crop identification feature optimization method and system for plain and hilly areas

Publications (1)

Publication Number Publication Date
CN116503681A true CN116503681A (en) 2023-07-28

Family

ID=87319680

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310453932.0A Pending CN116503681A (en) 2023-04-20 2023-04-20 Crop identification feature optimization method and system for plain and hilly areas

Country Status (1)

Country Link
CN (1) CN116503681A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117079143A (en) * 2023-10-16 2023-11-17 南京佳格耕耘科技有限公司 Farmland dynamic monitoring system based on remote sensing data

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117079143A (en) * 2023-10-16 2023-11-17 南京佳格耕耘科技有限公司 Farmland dynamic monitoring system based on remote sensing data

Similar Documents

Publication Publication Date Title
Belgiu et al. Sentinel-2 cropland mapping using pixel-based and object-based time-weighted dynamic time warping analysis
Halme et al. Utility of hyperspectral compared to multispectral remote sensing data in estimating forest biomass and structure variables in Finnish boreal forest
Li et al. Multi-LUTs method for canopy nitrogen density estimation in winter wheat by field and UAV hyperspectral
CN111598019B (en) Crop type and planting mode identification method based on multi-source remote sensing data
Fang et al. Retrieving leaf area index using a genetic algorithm with a canopy radiative transfer model
CN111898503B (en) Crop identification method and system based on cloud coverage remote sensing image and deep learning
CN112861435B (en) Mangrove quality remote sensing inversion method and intelligent terminal
CN111462223B (en) Sentinel-2 image-based planting area identification method for soybeans and corns in Jianghuai region
CN112364681B (en) Vegetation coverage estimation method and device based on two-dimensional table
Luo et al. Combining spectral and texture features for estimating leaf area index and biomass of maize using Sentinel-1/2, and Landsat-8 data
Jiao et al. Improving MODIS land cover classification by combining MODIS spectral and angular signatures in a Canadian boreal forest
CN113011235A (en) Multi-source remote sensing grassland and livestock balance evaluation method based on CPUGPU heterogeneous platform
CN116503681A (en) Crop identification feature optimization method and system for plain and hilly areas
CN115147721A (en) Remote sensing inversion method and device for forest accumulation
Zhu et al. Identification of soybean based on Sentinel-1/2 SAR and MSI imagery under a complex planting structure
Rakuasa et al. Analysis of Vegetation Index in Ambon City Using Sentinel-2 Satellite Image Data with Normalized Difference Vegetation Index (NDVI) Method based on Google Earth Engine
Heremans et al. Efficient collection of training data for sub-pixel land cover classification using neural networks
Yuan et al. Research on rice leaf area index estimation based on fusion of texture and spectral information
CN114494909A (en) Method and system for generating spatial distribution diagram of soybean growing season
CN115719453B (en) Deep learning-based remote sensing extraction method for rice planting structure
CN114332546B (en) Large-scale migration learning crop classification method and system based on phenological matching strategy
CN114782835A (en) Crop lodging area proportion detection method and device
Dadi Assessing the transferability of random forset and time-weighted dynamic time warping for agriculture mapping
Atzberger et al. Portability of neural nets modelling regional winter crop acreages using AVHRR time series
Reyes-Muñoz et al. Inferring global terrestrial carbon fluxes from the synergy of Sentinel 3 & 5P with Gaussian process hybrid models

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