CN117475313A - Large-scale winter wheat early-stage identification method based on automatic extraction of training samples - Google Patents
Large-scale winter wheat early-stage identification method based on automatic extraction of training samples Download PDFInfo
- Publication number
- CN117475313A CN117475313A CN202311495045.6A CN202311495045A CN117475313A CN 117475313 A CN117475313 A CN 117475313A CN 202311495045 A CN202311495045 A CN 202311495045A CN 117475313 A CN117475313 A CN 117475313A
- Authority
- CN
- China
- Prior art keywords
- winter
- winter wheat
- sentinel
- wheat
- ndvi
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 241000209140 Triticum Species 0.000 title claims abstract description 139
- 235000021307 Triticum Nutrition 0.000 title claims abstract description 139
- 238000012549 training Methods 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000000605 extraction Methods 0.000 title claims abstract description 32
- 235000006008 Brassica napus var napus Nutrition 0.000 claims abstract description 29
- 240000000385 Brassica napus var. napus Species 0.000 claims abstract description 29
- 238000013145 classification model Methods 0.000 claims abstract description 10
- 238000007637 random forest analysis Methods 0.000 claims description 16
- 230000011218 segmentation Effects 0.000 claims description 13
- 238000011160 research Methods 0.000 claims description 11
- 230000010287 polarization Effects 0.000 claims description 10
- 230000004927 fusion Effects 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000004519 manufacturing process Methods 0.000 claims description 8
- 238000003306 harvesting Methods 0.000 claims description 7
- 230000003287 optical effect Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000009331 sowing Methods 0.000 claims description 6
- 238000001228 spectrum Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 claims description 3
- 230000035558 fertility Effects 0.000 claims description 3
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 claims description 2
- 238000000692 Student's t-test Methods 0.000 claims description 2
- 230000000873 masking effect Effects 0.000 claims description 2
- 238000012353 t test Methods 0.000 claims description 2
- 230000017260 vegetative to reproductive phase transition of meristem Effects 0.000 claims description 2
- 238000011161 development Methods 0.000 claims 1
- 230000018109 developmental process Effects 0.000 claims 1
- 238000010899 nucleation Methods 0.000 claims 1
- 238000005070 sampling Methods 0.000 claims 1
- 239000000284 extract Substances 0.000 abstract description 6
- 238000012544 monitoring process Methods 0.000 abstract description 4
- 235000013339 cereals Nutrition 0.000 abstract description 2
- 238000011156 evaluation Methods 0.000 abstract 1
- 238000012360 testing method Methods 0.000 abstract 1
- 238000013508 migration Methods 0.000 description 6
- 230000005012 migration Effects 0.000 description 6
- 238000011835 investigation Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 4
- 230000015572 biosynthetic process Effects 0.000 description 3
- 238000003786 synthesis reaction Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000002310 reflectometry Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 240000002234 Allium sativum Species 0.000 description 1
- 240000005979 Hordeum vulgare Species 0.000 description 1
- 235000007340 Hordeum vulgare Nutrition 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012272 crop production Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 235000004611 garlic Nutrition 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000012502 risk assessment Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Mining
-
- 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
- G06V10/70—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
- G06V20/13—Satellite images
Abstract
The invention provides a large-scale winter wheat early-stage identification method based on automatic extraction of training samples, which comprises the following steps: acquiring Sentinel-1 and Sentinel-2 satellite images in a wheat growing season through a remote sensing cloud platform; generating a time sequence of Sentinel-1 and Sentinel-2 which are continuous in time and space; constructing a winter crop index WCI; extracting winter crop pixels; distinguishing winter wheat from winter rape; constructing a winter wheat classification model by combining the Sentinel-1/2 image of the current year; obtaining a winter wheat classification product in the current year based on the classification model; and (3) migrating the training model to a target year, and determining the earliest identifiable period of winter wheat by testing the influence of the relative characteristics on the classification accuracy in different time windows. The method can extract winter wheat timely and accurately, and has great application potential in aspects of winter wheat planting area monitoring, yield prediction, grain safety evaluation and the like.
Description
Technical Field
The invention belongs to the field of agricultural condition remote sensing monitoring, and relates to a large-scale winter wheat early-stage identification method based on automatic extraction of training samples.
Background
The crop type remote sensing classified product can provide accurate crop space distribution and planting area information, is a basic base map for regional scale crop growth monitoring and productivity prediction, and is important for grain security risk assessment. Due to factors such as field breaking, complex planting modes and the like, the problems of low precision, low efficiency, poor universality and the like of the methods generally exist, and large-scale and high-precision crop classification products are still deficient. The existing crop classification product production based on remote sensing technology uses machine learning methods, and the methods seriously depend on ground real data to train a machine learning classifier and construct a crop classification model. Generally, the acquisition of ground real data requires field point-by-point investigation, and has high cost and low efficiency of time, manpower and material resources. Therefore, developing a set of training data automatic extraction method is important for the automatic production of a large-scale crop classification product. In addition, the published release time of most crop classification products exists with serious hysteresis, and the requirement of accurate management of crop production in the season is difficult to meet. How to perform early high-precision identification of crop classification products in a limited time based on limited images is also a problem to be solved.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a large-scale winter wheat early-stage identification method based on automatic extraction of training samples, which is characterized in that the spectrum difference of winter crops and other ground objects is clarified through time sequence optical Sentinel-2 data, winter crop index enhanced winter crop signals are constructed, and winter crop pixels are automatically extracted by using an Otsu threshold segmentation algorithm. Based on the polarization characteristics of Sentinel-1VH, winter wheat and winter rape are further distinguished, so that a large amount of high-quality winter wheat training samples are extracted. Finally, the automatic production of the distributed product of the winter wheat with the spatial resolution of 10 meters is realized based on a machine learning method, and the classification precision and the earliest identifiable period of the winter wheat in each period are defined based on a model migration method. The method overcomes the dependence of the traditional classification method on ground truth data, has high precision, efficiency and robustness, and can be widely applied to the rapid production and release of large-scale, high-precision and high-spatial-resolution winter wheat spatial distribution products.
The technical solution for realizing the purpose of the invention is as follows:
a large-scale winter wheat early-stage identification method based on automatic extraction of training samples comprises the following steps:
step 1: and acquiring Sentinel-1 and Sentinel-2 satellite images of the winter wheat in the growing season of the winter wheat which is mainly produced by China, and respectively preprocessing the images on a remote sensing cloud platform to generate time-space continuous Sentinel-1 and Sentinel-2 month time sequences.
Step 2: constructing winter crop index WCI based on Sentinel-2 time sequence and winter crop weather information, and extracting winter crop pixels, specifically comprising:
step 2-1: performing ground investigation and document retrieval to obtain the surface coverage type of a research area and crop fertility process information;
step 2-2: analyzing the difference of the weather of winter crops and non-winter crops based on the Sentinel-2NDVI time sequences of various surface coverage types;
step 2-3: based on the winter crop and the non-winter crop climate differences, winter crops are divided into a pre-growth stage (Early Growth Stage, EGS) and a post-growth stage (Late Growth Stage, LGS), wherein the pre-growth stage is a sowing stage to a heading stage, and the post-growth stage is a heading stage to a harvest stage. Further in NDVI time seriesSearching for sowing, heading and harvesting period of winter crops, extracting NDVI of corresponding growth period, and respectively recording as NDVI min1 ，NDVI max And NDVI min2 Then constructing a winter crop index WCI based on the indexes; in the step 2-3, the specific formula of the winter crop index WCI is as follows: wci=Δndvi EGS ×ΔNDVI LGS ＝(NDVI max -NDVI min1 )×(NDVI max -NDVI min2 )
Wherein Δndvi EGS And Δndvi LGS The increase and decrease of NDVI of winter crops in early and later growth stages are respectively min1 ，NDVI max And NDVI min2 NDVI values during sowing, heading and harvest of winter crops are represented, respectively.
Step 2-4: dividing a research area into grids of 1 degree multiplied by 1 degree, carrying out adaptive threshold segmentation on WCI (WCI) by the grids based on an Otsu threshold segmentation algorithm, and automatically extracting winter crop pixels; in the step 2-4, dividing a research area into grids of 1 degree multiplied by 1 degree, searching an optimal segmentation threshold OT based on an Otsu self-adaptive threshold segmentation algorithm, and carrying out grid-by-grid segmentation on WCI based on the OT so as to automatically extract winter crop pixels; the specific search formula of OT is:
σ 2 (t)＝P 0 (μ 0 -μ T ) 2 +P 1 (μ 1 -μ T ) 2
wherein sigma 2 (t) is the inter-class variance produced by dividing winter crops and other features based on a threshold t, P is the probability that undefined pixels belong to winter crops or other features, μ 0 ，μ 1 Sum mu T Is the average value of winter crops, other ground objects and all ground objects.
Step 2-5: further combining with a plough map layer of a ground cover product (FROM_GLC10), masking the extracted winter crop pixels to eliminate errors in the extracted winter crop pixels.
Step 3: based on the spectrum index and the backscattering difference of different winter crops in the Sentinel-1 and the Sentinel-2 time sequences, the method for distinguishing winter wheat from winter rape in winter crops specifically comprises the following steps:
step 3-1: extracting the spectrum indexes and the backscattering coefficients of winter wheat and winter rape based on the Sentinel-1 and Sentinel-2 time sequences according to ground survey data;
step 3-2: analyzing the difference of the spectral indexes and polarization bands of winter wheat and winter rape, determining sensitive bands and weather periods, and recording as SP t ；
Step 3-3: according to SP t Threshold value for distinguishing winter wheat from winter rape, if SP t <T is considered to be winter wheat pixels in winter crops, otherwise winter rape pixels.
Step 4: the winter rape and non-winter crop pixels are fused and defined as non-winter wheat pixels, then the winter wheat and the non-winter wheat pixels are sampled layer by grids, and 1000 pixels are randomly extracted from each grid as candidate training samples. And finally, further removing samples with insufficient representativeness from the candidate samples based on the space neighborhood analysis, wherein the rest is the final training samples.
Step 5: inputting the training samples obtained in the step 4 and independent Sentinel-1, sentinel-2 and fusion time sequence characteristics of the Sentinel-1 and the Sentinel-2 into a Random Forest (RF) classifier, constructing an RF classification model, and classifying winter wheat in 2020 and 2021 years, thereby verifying the reliability of an automatic training sample extraction method and the quality of the training samples, and determining the contribution of optical and radar characteristics to winter wheat classification. In combination with the automatically generated training data and the optimal features and RF classifier, a 10 meter spatial resolution winter wheat classification product was generated for study areas 2020 and 2021.
Step 6: and (3) migrating the RF classification model under the optimal characteristic combination in the step (5) to a target year (2021 and 2022), firstly generating a winter wheat classification product with the spatial resolution of 10 meters in the target year based on the same characteristics, and evaluating the reliability of the model migration method according to the accuracy of the winter wheat classification product. And then further classifying the winter wheat by combining the image features of the target years in different periods, evaluating the classification precision of the winter wheat in different periods, and finally determining the earliest identifiable period of the winter wheat.
Compared with the prior art, the technical scheme provided by the invention has the following technical effects:
1. according to the large-scale winter wheat early recognition method based on automatic extraction of the training samples, the training samples are extracted by fusion of the weather features and the multi-source remote sensing data, and the method is high in efficiency and strong in universality;
2. according to the large-scale winter wheat early-stage identification method based on automatic extraction of the training samples, the automatic extraction of the training data and the machine learning method are adopted on the remote sensing cloud platform, and the large-scale high-spatial-resolution winter wheat spatial distribution product production is carried out.
3. According to the large-scale winter wheat early-stage identification method based on automatic extraction of training samples, the production of the winter wheat spatial distribution products of the target year is carried out based on the historical classification model and model migration, the earliest identifiable period of winter wheat is definitely the heading period based on the accuracy of the winter wheat classification products of different periods, and the method is high in reliability, strong in timeliness and easy to popularize.
Drawings
FIG. 1 is a technical roadmap of the large-scale winter wheat early recognition method based on automatic extraction of training samples of the present invention;
FIG. 2 is a schematic diagram of winter crop index construction based on a training sample automated extraction large scale winter wheat early recognition method of the present invention;
FIG. 3 is a graph of different optical and radar timing characteristics changes for the large scale winter wheat early recognition method based on automated extraction of training samples of the present invention, wherein (A), (B), (C) and (D) are NDVI, NDYI, VV and VH characteristics, respectively;
FIG. 4 is a graph showing comparison of classification accuracy of winter wheat in year 2020 and 2021 of different classification characteristics and classifiers of the large-scale early recognition method of winter wheat based on automatic extraction of training samples according to the present invention, wherein (A) & (E), (B) & (F), (C) & (G) and (D) & (H) represent OA, F1, UA and PA, green and orange represent classification accuracy of winter wheat based on Sentinel-2 characteristics and RF classifier, and based on Sentinel-1 and Sentinel-2 fusion characteristics and RF classifier, respectively;
FIG. 5 is a graph showing comparison of classification accuracy of winter wheat in each of 2021 and 2022 years based on RF model migration of a large-scale early recognition method of winter wheat by automatic extraction of training samples, wherein (A) & (E), (B) & (F), (C) & (G) and (D) & (H) represent OA, F1, UA and PA, respectively, green and orange represent classification accuracy of winter wheat based on Sentinel-2 feature and RF classifier, and on Sentinel-1 and Sentinel-2 fusion feature and RF classifier, respectively;
FIG. 6 is a schematic diagram showing the results of consistency verification of remote sensing estimated areas of large-scale winter wheat and agricultural statistics based on an automatic extraction of training samples in 2020-2022 year winter wheat and agricultural statistics, wherein (A) and (B) are results of consistency verification of remote sensing estimated areas of 2020 and 2021 year winter wheat and agricultural statistics based on an automatic extraction of samples in provincial scale, (C) and (D) are results of consistency verification of remote sensing estimated areas of 2021 and 2022 year winter wheat and agricultural statistics based on model migration in provincial scale, and (E) and (F), (G) and (H) are corresponding results of consistency verification in municipal scale;
FIG. 7 is a graph comparing the classification results of winter wheat with those of other similar products based on the large-scale early recognition method of winter wheat by automatically extracting training samples, wherein the first column is a Sentinel-2 image, the second column is a ChinaWheat10 result of the invention, and the third, fourth and fifth columns are the results of other similar products (TWDTW, PTDTW, chinaCP) respectively;
FIG. 8 is a comparison chart of classification accuracy of winter wheat at different periods under different characteristics of the large-scale early recognition method of winter wheat based on automatic extraction of training samples, wherein a blue fold line is the accuracy corresponding to a single time phase Sentinel-2 characteristic (S2), a green fold line is the accuracy corresponding to a single time phase Sentinel-2 and a Sentinel-1 fusion characteristic (S2+S1), a red fold line is the accuracy corresponding to a multi-time phase Sentinel-2 and a Sentinel-1 fusion characteristic (S2+S1+T), and a light blue background frame represents the earliest recognizable period of winter wheat.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only for explaining the present invention and are not to be construed as limiting the present invention.
The practice of the present invention is based on Sentinel-1 and Sentinel-2 satellite images, with image parameters and the features used being shown in Table 1.
Table 1 satellite image parameters and related features for research
The embodiment is developed in the main production provinces of winter wheat in China (Huang-Huai-Hai: jiangsu, anhui, hubei, henan, shandong, hebei, shanxi and northwest areas: gansu, xinjiang and Sichuan basin: sichuan), and the coverage area of the areas is about 400 ten thousand km 2 The planting proportion of winter wheat accounts for over 96% of the planting area of winter wheat in China. Moreover, the automatic extraction of training samples is repeatedly carried out and verified in 2020 and 2021, and the early recognition of winter wheat is repeatedly carried out and verified in 2021 and 2022, so that the advancement, universality and robustness of the invention can be effectively verified.
The large-scale winter wheat early-stage identification method based on automatic extraction of training samples, as shown in fig. 1, mainly comprises image preprocessing, ground surface true data collection, automatic extraction of training samples, drawing and verification of winter wheat and monitoring of the earliest identifiable period, and specifically comprises the following steps:
step 1: satellite images of Sentinel-2 and Sentinel-1 in the growing season of winter wheat which is the main product of winter wheat in China are obtained, wherein images of Sentinel-2 with the scenes 65766, 65465 and 65305 are obtained in 2020, 2021 and 2022 respectively, and images of Sentinel-1 with the scenes 4955, 4826 and 4477 are obtained. The method comprises the steps of preprocessing Sentinel-2 and Sentinel-1 images on a remote sensing cloud platform, wherein the preprocessing step of Sentinel-2 comprises cloud removal, vegetation index calculation, month median synthesis and missing value filling, and the preprocessing step of Sentinel-1 mainly comprises denoising, polarization index calculation and month median synthesis. The Sentinel-2 cloud probability data set can represent the probability of cloud pollution of each pixel in the Sentinel-2 image, and cloud removal processing is firstly carried out on an original Sentinel-2 surface reflectivity image based on the data set to obtain a cloud-free Sentinel-2 image; then, the invention calculates three common vegetation indexes of NDVI, EVI and LSWI based on the Sentinel-2 original wave band to increase the sensitivity classification characteristic of winter wheat; further, screening all the Sentinel-2 images month by month, and fusing the Sentinel-2 images of each month by using a median synthesis algorithm so as to obtain month images; and finally, filling the vacant pixels on the moon image by using a linear interpolation method, thereby obtaining a space-time continuous Sentinel-2 month time sequence. For Sentinel-1, the invention firstly carries out denoising treatment on each scene of Sentinel-1 image based on a Refine Lee algorithm; then calculating two polarization characteristics of Ratio and Diff; and finally, combining the median value of the monthly Sentinel-1 image into a first period in the same Sentinel-2 processing mode, and generating a Sentinel-1 month time sequence.
Step 2: constructing winter crop index WCI based on Sentinel-2 time sequence and winter crop weather information, and extracting winter crop pixels, specifically comprising:
step 2-1: and carrying out ground investigation and document retrieval to obtain the surface coverage type of the research area and the crop fertility progress information. In winter, the main types of features in the research area include winter crops, forests (evergreen and deciduous), buildings, bodies of water and spring crops. Among them, winter crops mainly include winter wheat (about 91%) and winter rape (about 8%), and a small amount of garlic, barley, etc. (1%).
Step 2-2: based on the Sentinel-2NDVI time series of various types of surface coverage, the difference in the climate of winter crops and non-winter crops was analyzed. For non-vegetation ground features such as water bodies, buildings and the like, the NDVI of the non-vegetation ground features is basically stable in winter crop growing season, and the NDVI value is lower. The evergreen forest remains substantially stable despite its higher NDVI value. The fallen leaf forest can be reduced firstly and then increased in the winter crop growing season, the spring crop NDVI can be obviously increased at the end of the winter crop growing season, and the winter crop can be rapidly reduced after the rising in the whole growing season.
Step 2-3: winter crops are divided into a pre-growth stage (Early Growth Stage, EGS) and a post-growth stage (Late Growth Stage, LGS) based on the winter crop and non-winter crop climatic differences, wherein the pre-growth stage is defined as a sowing stage to a heading stage and the post-growth stage is defined as a heading stage to a harvesting stage (FIG. 2). From FIG. 2, it can be seen that the variation of NDVI of winter crops in the early and late growth stages is significantly higher than that of other features, and is therefore based on
The product of the NDVI variation amplitude of the EGS and the LGS can effectively enhance the difference between winter crops and other ground features. NDVI of the crop in winter are respectively at local minimum or maximum, and NDVI in corresponding growth period can be extracted by searching the time sequence of NDVI and respectively recorded as NDVI min1 ,NDVI max And NDVI min2 The winter crop index WCI is then calculated based on these indices.
Step 2-4: because of the large area coverage, there are approximately 400 hundred million pixels, and it is difficult to process the massive data set at a time even using cloud computing technology. Therefore, the invention divides the research area into grids of 1 degree multiplied by 1 degree, carries out self-adaptive threshold segmentation on the WCI by grids based on an Otsu threshold segmentation algorithm, and automatically extracts winter crop pixels.
Step 2-5: since the reflectivity of some pixels may be abnormal due to other factors, it is misidentified as a winter crop pixel. The invention further obtains the global 10 m spatial resolution surface coverage product FROM_GLC10, extracts a cultivated map layer FROM the product, then masks the extracted winter crop pixels based on the cultivated map layer, and eliminates possible errors in the extracted winter crop pixels.
Step 3: based on the spectrum index and the backscattering difference of different winter crops in the Sentinel-1 and the Sentinel-2 time sequences, the method for distinguishing winter wheat from winter rape in winter crops specifically comprises the following steps:
step 3-1: according to ground investigation data, the spectral indexes and the backscattering coefficients of winter wheat and winter rape are extracted based on the Sentinel-1 and Sentinel-2 time sequences. The invention extracts the most commonly used normalized vegetation index NDVI (fig. 3A) from the optical data, and the normalized yellow vegetation index NDYI (fig. 3B) sensitive to yellow, and extracts all polarization features VV (fig. 3C) and VH (fig. 3D) of the radar data.
Step 3-2: based on t-test statistical analysis, the difference of the spectrum indexes and polarization wave bands of winter wheat and winter rape is analyzed, the VH polarization difference of winter wheat and winter rape in the later flowering stage (4 months) of rape is found to be most obvious, and the trend can be seen in figure 3D.
Step 3-3: according to VH polarization threshold value, winter wheat and winter rape are distinguished, and 11734 and 11675 pure winter wheat and winter rape pixels are analyzed, so that VH= -16 can be distinguished most accurately. VH < -16 is considered to be winter wheat pixel in winter crops, otherwise winter rape pixel.
Step 4: the winter rape and non-winter crop pixels are fused and defined as non-winter wheat pixels, then the winter wheat and the non-winter wheat pixels are sampled layer by grids, and 1000 pixels are randomly extracted from each grid as candidate training samples. And finally, judging whether the candidate sample and the pixels around the candidate sample are of the same type or not based on a space neighborhood analysis technology, if so, considering that the sample is representative, otherwise, deleting, and finally, reserving the training sample with high quality and strong representativeness. Finally 189650 (winter wheat: 45313, others: 144337) and 190192 (winter wheat: 46148, others: 144044) training samples were obtained in 2020 and 2021, respectively.
Step 5: inputting the training samples obtained in the step 4 and independent Sentinel-1, sentinel-2 and Sentinel-1 and Sentinel-2 fusion time sequence characteristics into a Random Forest (RF) classifier, constructing an RF classification model, and classifying winter wheat in 2020 and 2021 years, thereby verifying the reliability of the automatic extraction method of the training samples and the quality of the training samples, and definitely determining the contribution of optical and radar characteristics to winter wheat classification. It can be seen from FIG. 4 that the accuracy of fusing the Sentinel-1 and Sentinel-2 data is significantly better than that based on the Sentinel-2 data alone, whether 2020 or 2021. In combination with the automatically generated training data and the optimal features and RF classifier, a 10 meter spatial resolution winter wheat classification product was generated for study areas 2020 and 2021. The total precision of the produced winter wheat classification products is over 94 percent (figures 4A and E), the correlation between the city level and the statistical data in the province is over 0.96 and 0.92 (figures 6A, B, E and F), and the wheat Tian Shibie is accurate (the second column of figure 7) and is obviously superior to other similar products (the third, fourth and fifth columns of figure 7).
Step 6: and (3) migrating the RF classification model trained by the optimal features (the fusion features of the Sentinel-2 and the Sentinel-1) in the step (5) to the target year (2021 and 2022), firstly generating a target year 10 m spatial resolution winter wheat classification product based on the same features, and evaluating the reliability of the model migration method according to the accuracy of the winter wheat classification product. The first 10 m spatial resolution of the 2021 and 2022 China winter wheat spatial distribution products has overall accuracy of more than 93% (figures 5A and E) and correlation with statistical data at province and city levels of more than 0.95 and 0.91 respectively (figures 6C, D, G and H). And then further classifying the winter wheat by combining image features of different periods of the target year, evaluating classification accuracy of the winter wheat in different periods, and determining the earliest identifiable period of the winter wheat based on the saturation point of the classification accuracy. FIG. 8 shows that the precision of winter wheat is over 80 percent after the overwintering period (2 months), the precision of the jointing period (3 months) is over 85 percent, and the heading period (4 months) can reach the highest precision (> 90 percent). The earliest identifiable period of winter wheat is the heading period.
Step 7: and verifying the classification precision of winter wheat based on different methods:
the method comprises the following steps: in 2020-2022, 1148 winter wheat samples and 581 non-winter wheat samples are obtained through ground investigation, 1775 winter wheat samples and 3994 non-winter wheat samples are obtained through visual interpretation of high-resolution images. Constructing a confusion matrix by ground survey data and visual interpretation data, and generating overall accuracy based on the confusion matrix
(overlay Accurcry, OA), F1 score (F1-score, F1), user precision (UA) and Producer's Accurcry, PA) index quantitatively evaluate the positional precision of the winter wheat classification product.
The second method is as follows: the invention further obtains statistical data of winter wheat planting area of province and city scale in the national and each province statistical bureau, compares the statistical data with the winter wheat area in the classification result, and is based on a decision coefficient (R 2 ) And quantitatively evaluating the area accuracy of the winter wheat classification products with respect to Root Mean Square Error (RMSE).
And a third method: the invention downloads PTDTW, TWDTW and ChinacP winter wheat classification products, and the recognition effect of the method provided by the invention on winter wheat is qualitatively evaluated from classification details by comparing the PTDTW, TWDTW and ChinacP winter wheat classification products.
While only a few embodiments of the present invention have been described, it should be noted that modifications could be made by those skilled in the art without departing from the principles of the present invention, which modifications are to be regarded as being within the scope of the invention.
Claims (7)
1. The large-scale winter wheat early-stage identification method based on automatic extraction of training samples is characterized by comprising the following steps of:
step 1: acquiring Sentinel-1 and Sentinel-2 satellite images in winter wheat growing season in a main production area of China winter wheat, respectively preprocessing the images on a remote sensing cloud platform, and generating time-space continuous Sentinel-1 and Sentinel-2 month time sequences;
step 2: constructing a winter crop index WCI based on the Sentinel-2 time sequence and the winter crop growth and development characteristics, and extracting winter crop pixels;
step 3: distinguishing winter wheat and winter rape in winter crops based on different winter crop spectral indexes and backscattering differences in the Sentinel-1 and Sentinel-2 time sequences;
step 4: the winter rape and non-winter crop pixels are fused and defined as non-winter wheat pixels, then the winter wheat and the non-winter wheat pixels are sampled layer by grids, and a plurality of pixels are randomly extracted from each grid to serve as candidate training samples; finally, based on the space neighborhood analysis, removing the sample with insufficient representativeness in the candidate samples, and determining the rest as a final training sample;
step 5: inputting the training samples obtained in the step 4, independent Sentinel-1, sentinel-2 and the fusion time sequence characteristics of the Sentinel-1 and the Sentinel-2 into a random forest RF classifier, constructing an RF classification model, and classifying winter wheat in a historical period, thereby verifying the reliability of an automatic training sample extraction method and the quality of the training samples, and determining the contribution of optical and radar characteristics to the classification of winter wheat; combining the automatically generated training data, the optimal characteristics and the RF classifier to generate a 10-meter spatial resolution winter wheat classified product in the historical period of the research area;
step 6: migrating the RF classification model under the optimal feature combination in the step 5 to a target year, and carrying out early recognition on winter wheat in the target year: firstly, generating a winter wheat classification product with the spatial resolution of 10 meters in the target year, then classifying the winter wheat by combining image features of different periods in the target year, evaluating the classification precision of the winter wheat in the different periods, and finally determining the earliest identifiable period of the winter wheat.
2. The method for early recognition of large-scale winter wheat based on automatic extraction of training samples as claimed in claim 2, wherein the step 2 specifically comprises:
step 2-1: acquiring the earth surface coverage type of a research area and crop fertility progress information;
step 2-2: analyzing the difference of the weather of winter crops and non-winter crops based on the Sentinel-2 normalized vegetation index NDVI time sequence of various surface coverage types;
step 2-3: dividing winter crops into an early growth stage EGS and a later growth stage LGS based on the weather difference of the winter crops and the non-winter crops, wherein the early growth stage is a seeding stage to a heading stage, and the later growth stage is a heading stage to a harvest stage; searching sowing period, heading period and harvesting period of winter crops on NDVI time sequence, extracting NDVI of corresponding growth period, and respectively recording as NDVI min1 ,NDVI max And NDVI min2 Then constructing a winter crop index WCI based on the indexes;
step 2-4: dividing a research area into grids of 1 degree multiplied by 1 degree, carrying out adaptive threshold segmentation on WCI (WCI) by the grids based on an Otsu adaptive threshold segmentation algorithm, and automatically extracting winter crop pixels;
step 2-5: and (3) masking the extracted winter crop pixels by combining the plough map layer of the ground surface covering product FROM_GLC10, so as to eliminate errors in the extracted winter crop pixels.
3. The method for early recognition of large-scale winter wheat based on automatic extraction of training samples according to claim 2, wherein in step 2-3, the specific formula of winter crop index WCI is:
WCI＝ΔNDVI EGS ×ΔNDVI LGS ＝(NDVI max -NDVI min1 )×(NDVI max -NDVI min2 )
wherein Δndvi EGS And Δndvi LGS The increase and decrease of NDVI of winter crops in early and later growth stages are respectively min1 ,NDVI max And NDVI min2 NDVI values during sowing, heading and harvest of winter crops are represented, respectively.
4. The method for early recognition of large-scale winter wheat based on automatic extraction of training samples according to claim 1, wherein in the step 2-4, a research area is divided into grids of 1 degree by 1 degree, an optimal segmentation threshold value OT is searched based on an Otsu self-adaptive threshold segmentation algorithm, and grid-by-grid segmentation is performed on WCI based on the OT, so that winter crop pixels are automatically extracted; the specific search formula of OT is:
σ 2 (t)＝P 0 (μ 0 -μ T ) 2 +P 1 (μ 1 -μ T ) 2
wherein sigma 2 (t) dividing winter crops and other features based on a threshold value tThe resulting inter-class variance, P, is the probability that an undefined pixel belongs to winter crops or other features, μ 0 ,μ 1 Sum mu T Is the average value of winter crops, other ground objects and all ground objects.
5. The method for early recognition of large-scale winter wheat based on automatic extraction of training samples as claimed in claim 1, wherein the step 3 specifically comprises:
step 3-1: extracting the spectrum indexes and the backscattering coefficients of winter wheat and winter rape based on the Sentinel-1 and Sentinel-2 time sequences according to ground survey data;
step 3-2: analyzing the difference of the spectral indexes and polarization bands of winter wheat and winter rape, determining sensitive bands and weather periods, and recording as SP t ；
Step 3-3: according to SP t Threshold value for distinguishing winter wheat from winter rape, if SP t <T is considered to be winter wheat pixels in winter crops, otherwise winter rape pixels.
6. The method for identifying the early stage of the large-scale winter wheat based on the automatic extraction of training samples according to claim 1, wherein in the step 3, the threshold value of the backscattering coefficient of Sentinel-1VH is used for distinguishing winter rape from winter wheat, and the specific steps are as follows:
firstly, extracting moon NDVI, NDYI, VV and VH characteristic time sequences according to winter rape and winter wheat ground survey data;
then, based on t-test and time sequence characteristic analysis, optimally distinguishing the period and the characteristic of winter rape from winter wheat, and finally selecting VH polarization in the later flowering stage of rape for distinguishing winter wheat from winter rape;
and finally, determining the optimal distinguishing threshold value of winter rape and winter wheat by combining the pixel distribution frequency diagram, and realizing the accurate distinguishing of winter wheat and winter rape pixels.
7. The method for early recognition of large-scale winter wheat based on automatic extraction of training samples according to claim 1, wherein in step 4, a hierarchical random sampling method is used to generate training samples grid by grid, and then errors in the training samples are further reduced according to a spatial neighborhood analysis technology, so that quality and representativeness of the training samples are enhanced.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311495045.6A CN117475313A (en) | 2023-11-10 | 2023-11-10 | Large-scale winter wheat early-stage identification method based on automatic extraction of training samples |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311495045.6A CN117475313A (en) | 2023-11-10 | 2023-11-10 | Large-scale winter wheat early-stage identification method based on automatic extraction of training samples |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117475313A true CN117475313A (en) | 2024-01-30 |
Family
ID=89627299
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311495045.6A Pending CN117475313A (en) | 2023-11-10 | 2023-11-10 | Large-scale winter wheat early-stage identification method based on automatic extraction of training samples |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117475313A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117851726A (en) * | 2024-03-05 | 2024-04-09 | 山东科技大学 | Method for calculating leaf age by using winter wheat accumulated temperature |
-
2023
- 2023-11-10 CN CN202311495045.6A patent/CN117475313A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117851726A (en) * | 2024-03-05 | 2024-04-09 | 山东科技大学 | Method for calculating leaf age by using winter wheat accumulated temperature |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109345555B (en) | Method for identifying rice based on multi-temporal multi-source remote sensing data | |
| Wang et al. | A combined GLAS and MODIS estimation of the global distribution of mean forest canopy height | |
| CN109948596B (en) | Method for identifying rice and extracting planting area based on vegetation index model | |
| Xie et al. | Remote sensing imagery in vegetation mapping: a review | |
| Chen et al. | Automatic mapping of planting year for tree crops with Landsat satellite time series stacks | |
| Razak et al. | Mapping rubber trees based on phenological analysis of Landsat time series data-sets | |
| CN112183209A (en) | Regional crop classification method and system based on multi-dimensional feature fusion | |
| CN111738066B (en) | Grid late rice sheath blight disease habitat evaluation method integrating multisource remote sensing information | |
| CN109063660B (en) | Crop identification method based on multispectral satellite image | |
| CN113505635A (en) | Method and device for identifying winter wheat and garlic mixed planting area based on optics and radar | |
| CN117475313A (en) | Large-scale winter wheat early-stage identification method based on automatic extraction of training samples | |
| CN111007013B (en) | Crop rotation fallow remote sensing monitoring method and device for northeast cold region | |
| CN116645603A (en) | Soybean planting area identification and area measurement method | |
| CN115953683A (en) | Method for detecting hyperspectral change through learning of small samples across heterogeneous domains based on bidirectional generation | |
| CN115641504A (en) | Automatic remote sensing extraction method for field boundary based on crop phenological characteristics and decision tree model | |
| CN105930863A (en) | Determination method for spectral band setting of satellite camera | |
| Makhamreh | Derivation of vegetation density and land-use type pattern in mountain regions of Jordan using multi-seasonal SPOT images | |
| Zhao et al. | Cropland abandonment mapping at sub-pixel scales using crop phenological information and MODIS time-series images | |
| Roy et al. | Comparative analysis of object based and pixel based classification for mapping of mango orchards in Sitapur district of Uttar Pradesh | |
| Sreelekha et al. | Accuracy assessment of supervised and unsupervised classification using NOAA data in Andhra Pradesh region | |
| Guo et al. | Evaluating automatic road detection across a large aerial imagery collection | |
| WO2023131949A1 (en) | A versatile crop yield estimator | |
| CN115797501A (en) | Time-series forest age mapping method combining forest disturbance and recovery events | |
| Liu et al. | Assessment of generalized allometric models for aboveground biomass estimation: A case study in Australia | |
| Yang et al. | Digital Soil Mapping Based on Fine Temporal Resolution Landsat Data Produced by Spatio-temporal Fusion |
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 |