CN115719453A - Rice planting structure remote sensing extraction method based on deep learning - Google Patents
Rice planting structure remote sensing extraction method based on deep learning Download PDFInfo
- Publication number
- CN115719453A CN115719453A CN202310025855.9A CN202310025855A CN115719453A CN 115719453 A CN115719453 A CN 115719453A CN 202310025855 A CN202310025855 A CN 202310025855A CN 115719453 A CN115719453 A CN 115719453A
- Authority
- CN
- China
- Prior art keywords
- rice
- remote sensing
- machine learning
- rice planting
- data
- 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.)
- Granted
Links
Images
Classifications
-
- 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
- Image Analysis (AREA)
Abstract
The invention discloses a rice planting structure remote sensing extraction method based on deep learning, and relates to the technical field of remote sensing classification. Because the method synchronously trains and tests various machine learning models, and organically combines the various machine learning models by utilizing a scoring strategy to judge and identify the rice fine classification which is difficult to distinguish by a single model, the ground features distinguished by the method are very fine, and the method can effectively and finely classify various different rice planting structure types of 'same object and same spectrum'.
Description
Technical Field
The invention relates to the technical field of remote sensing classification, in particular to a remote sensing extraction method of a rice planting structure based on deep learning.
Background
The rice is the staple grain crop with the largest planting area and the widest coverage range in China, and the accurate extraction of the rice planting structure has very important significance for the prediction of the rice yield. Under the combined action of various factors such as climate change and the like, the planting structure of regional rice is continuously changed. The rice can be divided into double-cropping rice, single-cropping late rice, middle rice and regenerated rice according to the difference of rice cropping systems. Wherein the double cropping rice, the single cropping late rice, the middle rice and the regeneration rice are used as the rice planting structure types and are in the third level subdivision type of farmland-rice type. These subdivided rice types are rice of the "same species and same spectrum", and the distinction is made only in the rice cropping system and the cultivation method, and it is more difficult to distinguish the ratoon rice from the double cropping rice. The remote sensing technology has the advantages of wide coverage range, short detection period, strong current situation, low cost and the like, becomes an important means for extracting the rice planting structure, and can accurately obtain the geographic spatial distribution information of large-area rice planting in real time by utilizing the remote sensing technology. The remote sensing technologies mainly comprise a single-scene spectrum image method, a phenological feature method based on time sequence spectrum and a machine learning classification algorithm, and the defects of the remote sensing technologies for extracting the rice planting structure are as follows:
the single-scene spectral imaging method is difficult to overcome the problems of foreign matter co-spectrum, same thing and different spectrum and the like, the classification result often has a salt and pepper phenomenon, the crop identification accuracy is influenced, and the conditions of large amount of wrong classification, missing classification and the like of the planting area exist.
The phenological characteristic method based on the time sequence spectrum adopts image data in the complete growth cycle of crops, solves the problems of foreign matter and spectrum identity and the like of a single-scene spectral image method, but carries out rice identification through a spectral index threshold value, the classification threshold value depends on manual intervention, subjective repeated deduction is needed for revision, and the classification precision of the spectral confusion region is still not high.
The machine learning classification algorithm has strong advantages in the aspect of processing high-dimensional data and redundant data, does not depend on manual intervention, and can remarkably improve the operation efficiency and classification accuracy compared with the traditional method. The existing machine learning classification algorithm comprises a Bayes classification algorithm, a random forest classification algorithm, a support vector machine classification algorithm, a decision tree classification algorithm, a K-nearest neighbor classification algorithm and the like.
However, the existing single machine learning classification algorithm, such as the random forest algorithm, still has difficulty in obtaining high-precision results or cannot be directly distinguished when distinguishing the subdivided rice types of the 'same-object same-spectrum'. Therefore, the problem of how to finely classify a plurality of different rice subdivision types by using a plurality of classification algorithms needs to be solved.
Disclosure of Invention
The method aims to solve the problem of how to combine various classification algorithms to finely classify various different rice subdivision types of 'same-species same-spectrum'.
In order to solve the technical problems, the invention provides the following technical scheme:
a rice planting structure remote sensing extraction method based on deep learning comprises the following steps:
acquiring remote sensing data of a target planting area, and forming an NDVI time sequence data set based on an NDVI model and the remote sensing data of the target planting area;
acquiring field sample point data of a rice planting area, and assigning an NDVI time sequence data set to each pixel point of remote sensing data of a target planting area based on the field sample point data to obtain rice planting sample points; the rice planting area is a rice planting area in the target planting area;
training and testing different machine learning algorithms by taking the rice planting sample points as input quantity and taking the on-site sample point data of a rice planting area as output quantity to obtain the verification precision of each machine learning algorithm;
acquiring remote sensing data of a target area to be detected, selecting a plurality of machine learning algorithms with standard verification precision to enter a scoring strategy, and performing rice type division on target pixels in the remote sensing data of the target area to be detected by the aid of the plurality of machine learning algorithms entering the scoring strategy respectively, wherein each machine learning algorithm correspondingly outputs one rice type;
and inputting the verification precision of each machine learning algorithm and the corresponding output rice type based on the scoring strategy to obtain the rice type of the target pixel in the remote sensing data of the target area to be detected. Preferably, the scoring strategy is specifically: after multiple machine learning algorithms enter a scoring strategy, the verification precision of each machine learning algorithm is used as a weight value to be assigned to the corresponding output rice type, the weight values with the same rice type under the same target pixel are summed to obtain a comprehensive assigned value, finally, the comprehensive assigned values of different rice types are compared, and the rice type with the maximum comprehensive assigned value is the rice type of the target pixel.
Preferably, the specific method for training and testing machine learning algorithm, which takes the rice planting sample points as input quantity and the solid sample point data of the rice planting area as output quantity, comprises the following steps:
dividing the rice planting sample points into a training set and a testing set, extracting an NDVI time sequence data set corresponding to the rice planting sample points as an input quantity, using the on-site sample point data of a rice planting area as an output quantity, training each machine learning algorithm by using the training set, calling a confusion matrix to perform quantitative evaluation on the precision of each machine learning algorithm, and verifying the precision by using the testing set.
Preferably, the specific method of forming the NDVI time series dataset from the normalized vegetation index within the target planting area is:
and calculating the NDVI month value according to the maximum value of the normalized vegetation index in the month based on the normalized vegetation index, and forming an NDVI time sequence data set according to the NDVI month value.
Preferably, the remote sensing data and the remote sensing data to be classified are satellite image data which are subjected to orthorectification, geometric precision rectification and atmospheric rectification.
Preferably, the machine learning algorithm includes, but is not limited to, a K-nearest neighbor classification algorithm, a decision tree classification algorithm, a naive bayes classification algorithm, a logistic regression classification algorithm, a support vector machine classification algorithm, a random forest classification algorithm.
The remote sensing extraction method of the rice planting structure aims at the rice type which is a rice subdivision type with different rice cropping systems or cultivation modes.
Further, the rice subdivision type includes ratoon rice.
The invention also provides a remote sensing extraction system of the rice planting structure based on deep learning, which is used for identifying the rice type of the remote sensing data of the target area to be detected by applying the remote sensing extraction method of the rice planting structure based on deep learning and comprises the following components:
the remote sensing data processing module: the device is used for calculating and processing the multi-temporal satellite image data to form an NDVI time sequence data set and outputting the NDVI time sequence data set to the storage module;
a storage module: the device is used for receiving and storing the NDVI time sequence data set and the on-site sample point data to form and store rice planting sample points; outputting the rice planting sample point data to a training module;
a training module: extracting NDVI month value data corresponding to the rice planting sample points in the storage module, inputting the NDVI month value data into a training machine learning classification algorithm, and outputting the verification precision of the machine learning classification algorithm to the identification module;
an identification module: and receiving and outputting the verification precision of the machine learning classification algorithm, and calculating and outputting the rice type of each pixel in the remote sensing data of the target area to be detected by using a scoring strategy.
The invention provides a computer storage medium, wherein a computer program is stored on the computer readable storage medium, and when the computer program is executed by a processor, the processor executes the steps of the remote sensing extraction method of the rice planting structure based on deep learning.
The invention has the following beneficial effects:
1. the method comprises the steps of forming rice planting sample points by combining processed remote sensing data with field sample points, simultaneously training and verifying the precision of a plurality of machine learning models by using the rice planting sample points, outputting the precision of the plurality of machine learning models to a scoring strategy, and identifying and judging the rice type of pixels through the scoring strategy. Because the method synchronously trains and tests a plurality of machine learning models, and organically combines the plurality of machine learning models by utilizing a scoring strategy to judge and identify rice subdivision types which are difficult to distinguish by a single model, the ground features distinguished by the method are very fine, and the method can finely classify a plurality of different rice subdivision types with the same object and the same spectrum.
2. The scoring strategy of the invention does not simply compare the precision of various machine learning models to output the rice types, but creatively assigns the precision of different machine learning models to the corresponding rice types as the weight, then combines the assigned values of the same rice types, compares the comprehensive assigned values on the basis and outputs the rice types. The output models of various machine learning methods are fully called in the assigning, combining and comparing processes, the various machine learning methods are organically combined to identify the rice types of the pixels, the identification accuracy is greatly improved, the defect of a single model is avoided, the advantages of the various models are complemented, and therefore the rice subdivision types can be accurately classified.
Drawings
Fig. 1 is a flowchart of a K-nearest neighbor classification algorithm for classifying rice segment types according to an embodiment of the present application.
Fig. 2 is a flowchart of a decision tree classification algorithm for classifying rice segment types according to an embodiment of the present disclosure.
Fig. 3 is a flow chart of a random forest for classifying rice subdivision types according to an embodiment of the present application.
Fig. 4 is a flowchart of a support vector machine for classifying rice segment types according to an embodiment of the present disclosure.
Fig. 5 is a general technical route diagram of the present invention.
Fig. 6 is a field sample point distribution plot.
FIG. 7 is a statistical chart of rice phenological stages for 4 rice types.
Detailed Description
The following describes specific embodiments of the present application in detail. It should be apparent that the described embodiments are only a few embodiments of the present application, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The regenerated rice is popularized and planted in recent years in China, and refers to a rice planting mode in which after first-season rice is harvested, axillary buds on rice stakes germinate and grow, and the rice is spilt and fruited through cultivation management measures, so that one season is harvested again. Compared with single cropping rice, the method for planting the ratoon rice can fully utilize the photo-thermal resource, and improve the multiple cropping index of the rice field and the rice yield; compared with double cropping rice, the method saves labor and fertilizer for planting the ratoon rice and has higher economic benefit. As a new rice cultivation mode, the traditional classification algorithm cannot adapt to the newly added rice type subdivision requirement, and the whole rice planting structure cannot be accurately subdivided.
In this embodiment, the remote sensing extraction of the rice planting structure is performed by taking four rice types of double cropping rice, single cropping late rice, medium rice and ratoon rice as examples, and the specific method comprises the following steps:
1. and acquiring remote sensing data of the target planting area for training and verifying the machine learning algorithm.
The sentry No. 2 satellite comprises two sentry 2A (Sentinel-2A) and sentry 2B (Sentinel-2B) satellites, the orbits of the two satellites are different by 180 degrees, and the image acquisition time resolution of the two satellites as a group is 5 days. The height of a multispectral imager (MSI) carried by a sentinel No. 2 satellite is 786 km, the width of the multispectral imager reaches 290 km, 13 spectral bands from visible light, near infrared to short wave infrared are covered, the spatial resolution is different along with the specific bands of 10 m, 20 m and 60 m, and the specific bands are shown in a table 1.
Watch (CN)
Sentinel 2 satellite waveband attribute
The image and the processing of the satellite remote sensing are both carried out on a GEE (Google Earth Engine) geographic remote sensing cloud computing platform.
2. Forming an NDVI time sequence data set based on the NDVI model and the remote sensing data of the target planting area;
in this embodiment, the image data of the sentinel 2 satellite of a certain year, which has undergone orthorectification, geometric accuracy rectification and atmospheric rectification, is called by a code, and NDVI (normalized vegetation index) is calculated by band operation, the NDVI calculation formula is as follows,
in the formula: NIR refers to near infrared Band spectrum, is data Band 4 of a sentinel 2, and has the Band spatial resolution of 10 m; RED refers to RED Band spectrum, which is sentinel 2 data Band 8, and the Band spatial resolution is 10 m.
Based on the normalized vegetation index, calculating the maximum value of the normalized vegetation index in a month to obtain an NDVI month value, and forming an NDVI time sequence data set by the NDVI month value, wherein the operation method in the example is as follows:
and taking the maximum value of the NDVI of the multi-scene images in the month as the NDVI data of the month, generating month-by-month NDVI data of 12 months in the year, and forming an NDVI time sequence data set which is input data of a machine learning method.
3. Rice planting sample points are obtained and used for training and verifying a machine learning algorithm. Acquiring field sample point data of a rice planting area, and assigning an NDVI time sequence data set to each pixel point of remote sensing data of a target planting area based on the field sample point data to obtain rice planting sample points; the rice planting area is a rice planting area in the target planting area;
in this embodiment, a field survey is performed on a rice planting area in a research area in the year to obtain solid sample point data, and the solid sample point distribution is shown in fig. 6. Finally, 1000 rice planting sample points of 4 rice types are obtained, and the phenological period of the 4 rice types is shown in figure 7.
The specific method for training and testing machine learning algorithm, which takes rice planting sample points as input quantity and takes on-site sample point data of a rice planting area as output quantity, comprises the following steps:
dividing the rice planting sample points into a training set and a testing set according to a certain proportion, extracting an NDVI time sequence data set corresponding to the rice planting sample points as an input quantity, using field sample point data of a rice planting area as an output quantity, training each machine learning algorithm by using the training set, calling a confusion matrix to carry out quantitative evaluation on the precision of each machine learning algorithm, and then verifying the precision by using the testing set to obtain the verification precision of each machine learning algorithm.
In this embodiment, monthly NDVI monthly data of 12 months in a year corresponding to 1000 rice planting sample points is extracted as an input amount and is introduced into various machine learning algorithms, a field investigation result is used as a rice type identification true value, and the 1000 rice planting sample points are set according to the following equation of 7:3, the training set is used for training the machine learning algorithm, the test set is used for verifying the precision of the machine learning algorithm, the machine learning algorithm trained in the embodiment comprises a K nearest neighbor classification algorithm (fig. 1), a decision tree classification algorithm (fig. 2), a random forest classification algorithm (fig. 3) and a support vector machine classification algorithm (fig. 4), and a confusion matrix is called to quantitatively evaluate the precision of the classification model, so that the verification precision of each machine learning algorithm is obtained.
4. And outputting the rice type of the target pixel to be detected based on the scoring strategy.
Acquiring remote sensing data of a target area to be detected, selecting multiple machine learning algorithms with standard verification precision to enter a scoring strategy, performing rice type division on target pixels in the remote sensing data of the target area to be detected by the multiple machine learning algorithms entering the scoring strategy respectively, and outputting one rice type correspondingly by each machine learning algorithm;
and inputting the verification precision of each machine learning algorithm and the corresponding output rice type based on the scoring strategy to obtain the rice type of the target pixel in the remote sensing data of the target area to be detected.
And entering an algorithm model with the precision higher than 80% in various machine learning algorithms into a scoring strategy. The score strategy is: after various machine learning algorithms enter a scoring strategy, the verification precision of each machine learning algorithm is used as a weight value to be assigned to the rice type correspondingly output by the machine learning algorithm, the weight values with the same rice type under the same target pixel are summed to obtain a comprehensive assigned value, finally, the comprehensive assigned values of different rice types are compared, and the rice type with the maximum comprehensive assigned value is the rice type of the target pixel. For example, the verification precision values of the 4 machine learning algorithms in table 2 are directly used as weight values to be assigned to the respective correspondingly divided rice types, the weight values of double-cropping rice, single-cropping late rice, middle rice and ratoon rice under the target pixel 1 are respectively summed to obtain comprehensive assigned values, and finally the comprehensive assigned values of the 4 rice types are compared, wherein the rice type with the maximum comprehensive assigned value is the rice type output by the target pixel. Then, the assigning, summing and comparing processes are respectively repeated for the pixels 2 to 6, as shown in tables 2 and 3:
TABLE 2 results of rice type classification for different pixels by four machine learning algorithms
TABLE 3 comprehensive assigned values and rice type output results
The 4 machine learning algorithms in table 2 trained and yielded validation accuracies of 85%, 87%, 89%, and 91%, respectively (shown in table 3). The rice type division is respectively carried out on different pixels (target pixels 1-6) to be divided by using four machine learning algorithms, the rice type of the same pixel is divided by using different machine learning algorithms, different division results may exist, but only one true value of the same pixel exists, and then a scoring strategy is needed to judge and output the rice type. According to a scoring strategy, 85%, 87%, 89% and 91% are respectively used as the weight scores of 85%, 87%, 89 and 91 to be assigned to corresponding rice types, the weight scores of the same rice types under the same pixel are summed, and the rice type with the largest assigned value is used as an output type, so that the rice type of the target pixel can be identified (table 3).
Certainly, the 4-class algorithm in this embodiment may also be replaced with other machine learning algorithms, where the machine learning algorithm includes, but is not limited to, K nearest neighbor, decision tree, naive bayes, logistic regression, support vector machine, random forest, and the like.
As shown in fig. 1 to 4, the specific technical flows and precisions of the four machine learning algorithms in the embodiment are as follows:
1. operating environment
1. Equipped with Pycharm 2019.3.3
2. Importing a relational library
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn import svm
from sklearn.metrics import classification_report, confusion_matrix, accuracy_score
import joblib
2. Importing data
1. Sample points: 1000, which are divided into double cropping rice (250), single cropping late rice (250), ratoon rice (250) and middle cropping rice (250), and are respectively assigned with 1, 2, 3 and 4;
2. characteristic variables: 12, moon ndvi (normalized vegetation index)
3. Method for producing a composite material
1. Training set (70%), verification set (30%) are divided
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1,test_size=0.30)
2. Machine learning algorithm
(1) Random Forest classification algorithm (Random Forest)
classifier=RandomForestClassifier(n_estimators=65,max_depth=5,max_features=None,random_state=1)
Number of trees in forest 1 n _ estimators: integer, optional (default = 10) 10,50,100.
Integer or None, optionally (default = None) maximum depth of tree 5,8,15.
lmax _ features maximum number of features per decision tree
If"auto",then sqrt(n_features)
If"sqrt",then sqrt(n_features)(same as"auto")
If"log2",then log2(n_features)
If None,then n_features
If "float",then float*n_features
int/None, it is necessary to control the random state to repeatedly exhibit the same result
(2) Support vector machine classification algorithm (SVM)
classifier=svm.SVC(kernel='rbf',gamma=0.8,decision_function_shape='ovr',C=0.95)
In the case of l kernel = 'linear', the classification effect is better as C is larger, but overfitting may occur (defaul C = 1).
kernel = 'rbf' (default), which is a gaussian kernel function, the smaller the gamma value is, the more continuous the classification interface is; the larger the gamma value, the more "scattered" the classification interface, the better the classification effect, but may be overfitting.
And when the decision _ function _ shape = 'oo', the problem of one v one classification is solved, namely the classes are divided between every two classes, and a multi-classification result is simulated by a two-classification method.
When the decision _ function _ shape = 'ovr', the problem is classified as one v rest, that is, one class is classified from other classes.
(3) Decision Tree classification algorithm (Decision Tree)
classifier = DecisionTreeClassifier()
(4) K-nearest neighbor algorithm classification
classifier=KNeighborsClassifier(n_neighbors=7,weights='uniform',algorithm='auto',leaf_size=50,p=2,metric='minkowski',metric_params=None,n_jobs=None)
Int, optional parameter (default 5) number of default neighbors used for kneighbors queries
l weights str or callable, optional parameters (default is 'uniform')
Uniform weight the weight of points in each neighbor area is the same.
Distance' the weight points are equal to the inverse of their distance. Using this function, the influence of the closer neighbors on the predicted point is greater. [
A user-defined method receives an array of distances and then returns an array of identical shapes and containing weights.
l algorithm { 'auto', 'ball _ tree', 'kd _ tree', 'brute' }, optional parameters (default is 'auto')
The ball tree is invented for overcoming the high latitude failure of the kd-tree, the construction process is to divide a sample space by a centroid C and a radius r, and each node is a hyper-sphere.
The 'kd _ Tree' constructs a tree-like data structure in which kd-trees, i.e., binary trees in the data structure, store data for quick retrieval. The tree constructed by median segmentation has a hyper-rectangle of each node, and is efficient when the dimension is less than 20.
'brute' uses brute force search, i.e. linear scan, when the training set is large, the computation is very time consuming
The 'auto' will select the most appropriate algorithm based on the contents of the incoming fit method.
l leaf _ size int, optional parameter (default 30)
The number of leaves into the BallTree or KDTree algorithm. The parameters can influence the speed of constructing and querying the BallTree or the KDTree and the size of a memory required for storing the BallTree or the KDTree. This optional parameter is selectively used as required depending on whether it is a problem or not.
Integer, optional parameters (default 2)
Hyper-parameters for Minkowski metric (Minkowski space).
p =1, corresponding to the use of manhattan distance (l 1);
p =2, corresponding to the use of euclidean distance (l 2);
for any p, the Minkowski space (l _ p) is used.
String or callable, defaulted to 'minkowski'
Distance matrix for tree. Minkowski space is defaulted, if used with p =2 is equivalent to using a standard Euclidean matrix.
l metric _ params, dit, optional parameters (default to None), other key parameters for matrix methods.
Int, an optional parameter (default 1), for searching for the number of tasks of the neighbor that can run in parallel.
4. Method accuracy comparison
TABLE 4 precision of different methods
5. Model saving and calling
(1) Model output
Model=joblib.dump(classifier,'sdClassifier.pkl')
(2) Model invocation
Model=joblib.load('sdClassifier.pkl')。
The process comprises 4 main parameter setting and calling functions of a machine learning classification algorithm, as shown in fig. 5, after multi-time phase satellite image data is obtained, preprocessing (such as orthorectification, geometric precision correction and atmospheric correction processing) is carried out on the multi-time phase satellite image data, band synthesis and spectral index (such as NDVI) calculation are carried out, a calculation result (namely an NDVI time sequence dataset) is combined with field visual interpretation crop species data (namely field sample point data) to form a irrigated area scale long-time sequence spectral index (namely a rice planting sample point), after the main parameters are respectively set by a plurality of machine learning algorithms, the irrigated area scale long-time sequence spectral index is used as an input quantity to carry out parallel training and verification on the plurality of machine learning algorithms, rice type division is carried out on target pixels in remote sensing data of a target area to be tested by the plurality of machine learning algorithms with trained precision, the only rice subdivision type output value of the target pixels is finally confirmed by a scoring strategy, and the rice subdivision type output values of the plurality of target pixels are used to form an irrigated area scale rice planting structure.
The application also provides a remote sensing extraction system of the rice planting structure based on deep learning, which comprises the following components:
the remote sensing data processing module: the device is used for calculating and processing the multi-temporal satellite image data to form an NDVI time sequence data set, and outputting the NDVI time sequence data set to a storage module;
a storage module: the device is used for receiving and storing the NDVI time sequence data set and the field sample point data to form and store rice planting sample points; outputting the rice planting sample point data to a training module;
a training module: extracting NDVI month value data corresponding to the rice planting sample points in the storage module, inputting the NDVI month value data into a training machine learning classification algorithm, and outputting the verification precision of the machine learning classification algorithm to the identification module;
an identification module: and receiving and outputting the verification precision of the machine learning classification algorithm, and calculating and outputting the rice type of each pixel in the remote sensing data of the target area to be detected by using a scoring strategy.
The invention also provides a computer storage medium, wherein a computer program is stored on the computer readable storage medium, and when the computer program is executed by a processor, the processor is enabled to execute the steps of the remote sensing extraction method of the rice planting structure based on deep learning.
The present application has been described in detail with reference to specific embodiments and illustrative examples, but the description is not intended to limit the application. Those skilled in the art will appreciate that various equivalent substitutions, modifications or improvements may be made to the embodiments and implementations thereof without departing from the spirit and scope of the present application, and are within the scope of the present application. The protection scope of this application is subject to the appended claims.
Claims (10)
1. A rice planting structure remote sensing extraction method based on deep learning is characterized by comprising the following steps:
acquiring remote sensing data of a target planting area, determining the NDVI in the target planting area based on an NDVI model and the remote sensing data of the target planting area, and forming an NDVI time sequence data set by the NDVI in the target planting area;
acquiring field sample point data of a rice planting area, and assigning an NDVI time sequence data set to each pixel point of remote sensing data of a target planting area based on the field sample point data to obtain rice planting sample points; the rice planting area is a rice planting area in the target planting area;
training and testing different machine learning algorithms by taking the rice planting sample points as input quantity and taking the on-site sample point data of a rice planting area as output quantity to obtain the verification precision of each machine learning algorithm;
acquiring remote sensing data of a target area to be detected, selecting multiple machine learning algorithms with standard verification precision to enter a scoring strategy, performing rice type division on target pixels in the remote sensing data of the target area to be detected by the multiple machine learning algorithms entering the scoring strategy respectively, and outputting one rice type correspondingly by each machine learning algorithm;
and inputting the verification precision of each machine learning algorithm and the corresponding output rice type based on the scoring strategy to obtain the rice type of the target pixel in the remote sensing data of the target area to be detected.
2. The deep learning-based rice planting structure remote sensing extraction method according to claim 1, wherein the scoring strategy specifically comprises: after multiple machine learning algorithms enter a scoring strategy, the verification precision of each machine learning algorithm is used as a weight value to be assigned to the corresponding output rice type, the weight values with the same rice type under the same target pixel are summed to obtain a comprehensive assigned value, finally, the comprehensive assigned values of different rice types are compared, and the rice type with the maximum comprehensive assigned value is the rice type of the target pixel.
3. The remote sensing extraction method of the rice planting structure based on the deep learning of claim 1, wherein the specific method of training and testing machine learning algorithm with rice planting sample points as input quantity and field sample point data of rice planting area as output quantity is as follows:
dividing the rice planting sample points into a training set and a testing set, extracting an NDVI time sequence data set corresponding to the rice planting sample points as input quantity, using on-site sample point data of a rice planting area as output quantity, training each machine learning algorithm by using the training set, calling a confusion matrix to perform quantitative evaluation on the precision of each machine learning algorithm, and verifying the precision by using the testing set.
4. The rice planting structure remote sensing extraction method based on deep learning of claim 1, wherein the specific method for forming the NDVI time series dataset from the normalized vegetation indexes in the target planting area is as follows:
and calculating the NDVI month value according to the maximum value of the normalized vegetation index in the month based on the normalized vegetation index, and forming an NDVI time sequence data set according to the NDVI month value.
5. The deep learning-based rice planting structure remote sensing extraction method according to claim 1, wherein the remote sensing data and the remote sensing data to be classified are satellite image data which have been subjected to orthorectification, geometric accuracy rectification and atmospheric correction.
6. The remote sensing extraction method of the rice planting structure based on the deep learning of claim 1, wherein the machine learning algorithm includes but is not limited to a K-nearest neighbor classification algorithm, a decision tree classification algorithm, a naive Bayes classification algorithm, a logistic regression classification algorithm, a support vector machine classification algorithm, and a random forest classification algorithm.
7. The deep learning-based rice planting structure remote sensing extraction method as claimed in any one of claims 1 to 6, wherein the rice type is a rice subdivision type with different rice cropping systems or cultivation modes.
8. The deep learning-based remote sensing extraction method for rice planting structures according to claim 7, wherein the rice subdivision types comprise ratoon rice.
9. A rice planting structure remote sensing extraction system based on deep learning is characterized in that the rice type identification is carried out on remote sensing data of a target area to be detected by applying the rice planting structure remote sensing extraction method based on deep learning of any one of claims 1 to 6, and the system comprises the following components:
the remote sensing data processing module: the device is used for calculating and processing the multi-temporal satellite image data to form an NDVI time sequence data set, and outputting the NDVI time sequence data set to a storage module;
a storage module: the device is used for receiving and storing the NDVI time sequence data set and the field sample point data to form and store rice planting sample points; outputting the rice planting sample point data to a training module;
a training module: extracting NDVI month value data corresponding to the rice planting sample points in the storage module, inputting the NDVI month value data into a training machine learning classification algorithm, and outputting the verification precision of the machine learning classification algorithm to the identification module;
an identification module: and receiving and outputting the verification precision of the machine learning classification algorithm, and calculating and outputting the rice type of each pixel in the remote sensing data of the target area to be detected by using a scoring strategy.
10. A computer storage medium, wherein a computer program is stored on the computer readable storage medium, and when executed by a processor, the computer program causes the processor to execute the steps of the deep learning based remote sensing extraction method for rice planting structure according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310025855.9A CN115719453B (en) | 2023-01-09 | 2023-01-09 | Deep learning-based remote sensing extraction method for rice planting structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310025855.9A CN115719453B (en) | 2023-01-09 | 2023-01-09 | Deep learning-based remote sensing extraction method for rice planting structure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115719453A true CN115719453A (en) | 2023-02-28 |
| CN115719453B CN115719453B (en) | 2023-08-04 |
Family
ID=85257894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310025855.9A Active CN115719453B (en) | 2023-01-09 | 2023-01-09 | Deep learning-based remote sensing extraction method for rice planting structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115719453B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116152673A (en) * | 2023-04-21 | 2023-05-23 | 杭州稻道农业科技有限公司 | Method and device for identifying ratoons by utilizing multi-time-phase high-resolution satellite images |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109345555A (en) * | 2018-10-15 | 2019-02-15 | 中科卫星应用德清研究院 | Rice, which is carried out, based on multidate multi- source Remote Sensing Data data knows method for distinguishing |
| CN112464977A (en) * | 2020-10-15 | 2021-03-09 | 深圳先进技术研究院 | Object classification method, computer equipment and storage medium |
| CN113033670A (en) * | 2021-03-29 | 2021-06-25 | 华南农业大学 | Method for extracting rice planting area based on Sentinel-2A/B data |
| CN114220022A (en) * | 2021-12-10 | 2022-03-22 | 中国科学院南京土壤研究所 | Remote sensing monitoring method for rice lodging based on satellite and unmanned aerial vehicle cooperative observation |
| CN114254964A (en) * | 2022-03-01 | 2022-03-29 | 北京市农林科学院信息技术研究中心 | Rice regional climate quality assessment method and system |
| CN114782838A (en) * | 2022-06-17 | 2022-07-22 | 中化现代农业有限公司 | Rice identification method and device, electronic equipment and storage medium |
| CN114926532A (en) * | 2022-04-26 | 2022-08-19 | 江苏大学 | On-line detection method and system for height of regenerated rice ear layer and harvester |
| CN114998742A (en) * | 2022-06-16 | 2022-09-02 | 天津市生态环境科学研究院(天津市环境规划院、天津市低碳发展研究中心) | Method for quickly identifying and extracting rice planting area in single-season rice planting area |
| CN116152673A (en) * | 2023-04-21 | 2023-05-23 | 杭州稻道农业科技有限公司 | Method and device for identifying ratoons by utilizing multi-time-phase high-resolution satellite images |
-
2023
- 2023-01-09 CN CN202310025855.9A patent/CN115719453B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109345555A (en) * | 2018-10-15 | 2019-02-15 | 中科卫星应用德清研究院 | Rice, which is carried out, based on multidate multi- source Remote Sensing Data data knows method for distinguishing |
| CN112464977A (en) * | 2020-10-15 | 2021-03-09 | 深圳先进技术研究院 | Object classification method, computer equipment and storage medium |
| CN113033670A (en) * | 2021-03-29 | 2021-06-25 | 华南农业大学 | Method for extracting rice planting area based on Sentinel-2A/B data |
| CN114220022A (en) * | 2021-12-10 | 2022-03-22 | 中国科学院南京土壤研究所 | Remote sensing monitoring method for rice lodging based on satellite and unmanned aerial vehicle cooperative observation |
| CN114254964A (en) * | 2022-03-01 | 2022-03-29 | 北京市农林科学院信息技术研究中心 | Rice regional climate quality assessment method and system |
| CN114926532A (en) * | 2022-04-26 | 2022-08-19 | 江苏大学 | On-line detection method and system for height of regenerated rice ear layer and harvester |
| CN114998742A (en) * | 2022-06-16 | 2022-09-02 | 天津市生态环境科学研究院(天津市环境规划院、天津市低碳发展研究中心) | Method for quickly identifying and extracting rice planting area in single-season rice planting area |
| CN114782838A (en) * | 2022-06-17 | 2022-07-22 | 中化现代农业有限公司 | Rice identification method and device, electronic equipment and storage medium |
| CN116152673A (en) * | 2023-04-21 | 2023-05-23 | 杭州稻道农业科技有限公司 | Method and device for identifying ratoons by utilizing multi-time-phase high-resolution satellite images |
Non-Patent Citations (2)
| Title |
|---|
| 张雨: "大宗作物种植模式遥感智能监测关键技术与方法研究——以水稻为例", 《中国优秀硕士学位论文全文数据库农业科技辑》, no. 02, pages 043 - 17 * |
| 汪荃 等: "基于Sentinel-2时序影像的水稻种植信息提取", vol. 61, no. 16, pages 175 - 181 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116152673A (en) * | 2023-04-21 | 2023-05-23 | 杭州稻道农业科技有限公司 | Method and device for identifying ratoons by utilizing multi-time-phase high-resolution satellite images |
| CN116152673B (en) * | 2023-04-21 | 2023-09-29 | 杭州稻道农业科技有限公司 | Method and device for identifying ratoons by utilizing multi-time-phase high-resolution satellite images |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115719453B (en) | 2023-08-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Turkoglu et al. | Crop mapping from image time series: Deep learning with multi-scale label hierarchies | |
| Keshtkar et al. | Land-cover classification and analysis of change using machine-learning classifiers and multi-temporal remote sensing imagery | |
| Löw et al. | Impact of feature selection on the accuracy and spatial uncertainty of per-field crop classification using support vector machines | |
| González-Yebra et al. | Methodological proposal to assess plastic greenhouses land cover change from the combination of archival aerial orthoimages and Landsat data | |
| Ok et al. | 2-D delineation of individual citrus trees from UAV-based dense photogrammetric surface models | |
| Peters et al. | Synergy of very high resolution optical and radar data for object-based olive grove mapping | |
| LUO et al. | Comparison of machine learning algorithms for mapping mango plantations based on Gaofen-1 imagery | |
| Luo et al. | Crop type mapping in the central part of the North China Plain using Sentinel-2 time series and machine learning | |
| CN115965812B (en) | Evaluation method for classification of unmanned aerial vehicle images on wetland vegetation species and land features | |
| Mazarire et al. | Exploring machine learning algorithms for mapping crop types in a heterogeneous agriculture landscape using Sentinel-2 data. A case study of Free State Province, South Africa | |
| She et al. | Identification and mapping of soybean and maize crops based on Sentinel-2 data | |
| CN115719453B (en) | Deep learning-based remote sensing extraction method for rice planting structure | |
| CN116129262A (en) | Cultivated land suitability evaluation method and system for suitable mechanized transformation | |
| Alburshaid et al. | Palm trees detection using the integration between gis and deep learning | |
| Tu et al. | A 30 m annual cropland dataset of China from 1986 to 2021 | |
| Qu et al. | Mapping large area tea plantations using progressive random forest and Google Earth Engine | |
| Verma et al. | A review on land cover classification techniques for major fruit crops in India-Present scenario and future aspects | |
| Laine | Crop identification with Sentinel-2 satellite imagery in Finland | |
| Dahiya et al. | Comparative analysis and implication of Hyperion hyperspectral and landsat-8 multispectral dataset in land classification | |
| Pawar et al. | A Novel Hybrid AI Federated ML/DL Models for Classification of Soil Components | |
| László et al. | A fully segment-based method for the classification of satellite images | |
| Selea et al. | AgriSen-A Dataset for Crop Classification | |
| Ijabs et al. | Bidirectional Long Short-Term Memory Networks for Automatic Crop Classification at Regional Scale using Tabular Remote Sensing Time Series. | |
| Rajeswari et al. | Classification based land use/land cover change detection through Landsat images | |
| Prins | Efficacy of machine learning and lidar data for crop type mapping |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |