CN109145885B - Remote sensing classification method and system for large-scale crops - Google Patents
Remote sensing classification method and system for large-scale crops Download PDFInfo
- Publication number
- CN109145885B CN109145885B CN201811183899.XA CN201811183899A CN109145885B CN 109145885 B CN109145885 B CN 109145885B CN 201811183899 A CN201811183899 A CN 201811183899A CN 109145885 B CN109145885 B CN 109145885B
- Authority
- CN
- China
- Prior art keywords
- crop
- model
- samples
- rsnet
- historical
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a remote sensing classification method and system for large-scale crops. The method and the system establish a migration remote sensing network RSNet model according to a residual error network model and a pyramid pooling network model, pre-train the migration RSNet model according to historical training samples in large-range crop marking samples in historical years, and establish a pre-training migration RSNet model; and fine adjustment is carried out on the pre-training model based on the randomly distributed and independently distributed present small-area high-precision agricultural crop marker samples, and then rapid automatic classification of large-scale crops is realized on the present image based on the fine adjustment model. The fine-tuning model for the existing small-area crop marking sample provided by the invention greatly improves the classification precision of large-scale crops on the whole, realizes the migration of the spatial scale of the crop marking sample, solves the problem that the marking sample is limited by a specific area, a specific image and a specific target in the traditional crop classification, and improves the classification precision and the classification efficiency of the large-scale crops.
Description
Technical Field
The invention relates to the technical field of crop classification, in particular to a remote sensing classification method and system for large-scale crops.
Background
Efficient and accurate large-scale remote sensing image classification is a long-term remote sensing scientific problem. Crop remote sensing classification and spatial distribution monitoring thereof are main entry points of agricultural research, and are one of the key problems in the agricultural remote sensing popularization process. With the development of remote sensing data sources and remote sensing technologies, the traditional remote sensing image large-scale crop remote sensing classification method has made great progress in the aspects of classification accuracy and efficiency. According to different data sources, the traditional remote sensing large-scale crop classification method based on large-scale crop remote sensing can be divided into classification based on single-time-phase images, classification based on time-series images and classification based on multi-source compounding, and the remote sensing classification method for large-scale crops obtains larger achievement in a specific area. However, the regional limitation of the traditional feature extraction and training sample selection method restricts the application of the traditional large-scale crop remote sensing classification method in large-scale crop remote sensing identification.
In recent years, with the development of deep learning, different types of multi-level features in remote sensing images are automatically extracted, so that large-scale spatial distribution of crops is possible efficiently, and the deep learning method has gradually shown application potential in remote sensing image classification. However, one of the conditions for obtaining a high-accuracy model through deep learning is that a large number of marked samples are required, but in practical application, the acquisition of marked samples is time-consuming and labor-consuming, and the real-time rapid acquisition of marked training samples becomes a bottleneck problem of remote sensing large-scale application. In fact, the historically collected marking samples and the classification thematic maps can provide rich prior knowledge for new classification tasks, and meanwhile, according to the crop planting characteristics, the annual cultivated land range basically does not change, so that the spatial positions of the historically collected marking samples are very helpful for target classification of the present images. The traditional deep learning algorithm does not well apply the historical sample data with priori knowledge to the automatic extraction of new remote sensing information, so that the problem of sample data starvation often exists in the traditional remote sensing classification method for the large-scale crops of the large-scale crops, and the application of the large-scale automatic classification of the remote sensing images is limited.
Disclosure of Invention
The invention aims to provide a large-scale crop remote sensing classification method and system to improve classification precision and efficiency of large-scale remote sensing image classification and solve the problem of data hunger in the traditional large-scale crop remote sensing classification method.
In order to achieve the purpose, the invention provides the following scheme:
a remote sensing classification method for large-scale crops comprises the following steps:
obtaining a large-range crop marking sample and a present small-area crop marking sample in a historical year;
determining a historical training sample and a historical verification sample according to the wide-range agricultural marker samples of the historical years;
determining randomly distributed situation training samples, randomly distributed situation verification samples, independently distributed situation training samples and independently distributed situation verification samples according to the situation small-area crop marking samples;
acquiring a residual error network model and a pyramid pooling network model;
establishing a migration RSNet model according to the residual error network model and the pyramid pooling network model;
pre-training the migration RSNet model according to the historical training sample, and establishing a pre-training migration RSNet model;
fine tuning the pre-training migration RSNet model according to the randomly distributed situation training samples, and establishing a first fine tuning migration RSNet model;
fine tuning the pre-training migration RSNet model according to the independently distributed situation training samples, and establishing a second fine tuning migration RSNet model;
and carrying out provincial crop classification on the present image by adopting the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model.
Optionally, the acquiring of the historical year large-scale crop marking sample and the present small-area crop marking sample specifically includes:
acquiring historical multi-temporal time-sharing images and historical crop weather calendar data in a large-scale region of historical years; the large-range area is a province or above-province range area;
constructing a large-scale crop marking sample of the historical year according to the historical multi-temporal segmentation image and the historical crop weather calendar data;
acquiring present multi-temporal high-resolution images of small areas in the current year and current crop climate calendar data; the small areas are a plurality of counties;
and constructing the crop marking sample of the spot small area according to the spot multi-temporal height-divided image and the current crop calendar data.
Optionally, the constructing the large-scale crop marking sample in the historical year according to the historical multi-temporal median image and the historical crop calendar data specifically includes:
and classifying the crops by adopting a method of combining target change detection and a support vector machine according to the historical multi-temporal segmentation images and the historical crop calendar data, and constructing the large-range crop marking samples of the historical years.
Optionally, the constructing the crop marking sample of the present small region according to the present multi-temporal high-resolution image and the current crop phenological calendar data specifically includes:
performing image object segmentation on the present multi-temporal high-resolution image and the current crop climate data by adopting an object-oriented method to obtain a segmented image object of a designated area;
constructing spectral features and textural features of crops according to the segmented image objects in the designated area;
performing feature extraction and optimization on the spectral features and the textural features of the crops by adopting a random forest algorithm, performing high-precision crop classification by using a support vector machine classifier, and determining a high-resolution crop classification result;
and resampling the high-resolution crop classification result into a resolution which is the same as that of the medium-resolution crop marking sample by adopting a resampling method of a mode principle, and obtaining the present small-area crop marking sample.
Optionally, the determining a historical training sample and a historical verification sample according to the wide-range agricultural marker sample of the historical year specifically includes:
and selecting 85% of the historical year large-range crop marking samples as the historical training samples, and taking the rest 15% of the historical year large-range crop marking samples as the historical verification samples.
Optionally, the determining, according to the present small-area crop marker sample, a randomly distributed present training sample, a randomly distributed present verification sample, an independently distributed present training sample, and an independently distributed present verification sample specifically include:
and randomly selecting 85% of the crop mark samples in the present small area as the randomly distributed present training samples, and taking the rest 15% of the crop mark samples in the present small area as the randomly distributed present verification samples.
Selecting the crop marking samples of the present small areas in 85 percent of the designated areas adjacent to the small areas as the independently distributed present training samples, and using the remaining 15 percent of the crop marking samples of the present small areas as the independently distributed present verification samples.
Optionally, after the pre-training migration RSNet model, the first fine-tuning migration RSNet model, and the second fine-tuning migration RSNet model are adopted to classify the provincial crops of the present image, the method further includes:
performing precision evaluation on the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model by adopting a field investigation sample point; the indexes of the precision evaluation comprise overall classification precision, drawing precision, user precision, Kappa coefficient and F1 score representing weighted average of the drawing precision and the user precision.
A large-scale crop remote sensing classification system comprising:
the crop marking sample acquisition module is used for acquiring large-range crop marking samples and present small-area crop marking samples in historical years;
the historical sample determining module is used for determining a historical training sample and a historical verification sample according to the wide-range agricultural marker samples of the historical years;
the situation sample determination module is used for determining randomly distributed situation training samples, randomly distributed situation verification samples, independently distributed situation training samples and independently distributed situation verification samples according to the situation small-area crop marking samples;
the network model acquisition module is used for acquiring a residual network model and a pyramid pooling network model;
the migration RSNet model establishing module is used for establishing a migration RSNet model according to the residual error network model and the pyramid pooling network model;
the pre-training migration RSNet model establishing module is used for pre-training the migration RSNet model according to the historical training sample and establishing a pre-training migration RSNet model;
the first fine-tuning migration RSNet model building module is used for fine-tuning the pre-training migration RSNet model according to the randomly distributed situation training samples and building a first fine-tuning migration RSNet model;
the second fine tuning migration RSNet model is used for fine tuning the pre-training migration RSNet model according to the independently distributed situation training samples and establishing a second fine tuning migration RSNet model;
and the crop classification module is used for classifying provincial crops of the present images by adopting the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model.
Optionally, the history sample determining module specifically includes:
and the historical sample determining unit is used for selecting 85% of the historical year large-range crop marking samples as the historical training samples, and selecting the rest 15% of the historical year large-range crop marking samples as the historical verification samples.
Optionally, the present sample determination module specifically includes:
and the randomly distributed situation sample determining unit is used for randomly selecting 85% of the situation small-area crop marking samples as the randomly distributed situation training samples, and the rest 15% of the situation small-area crop marking samples as the randomly distributed situation verification samples.
And the independent distribution situation sample determination unit is used for selecting the situation small-area crop marking samples in 85 percent of the designated areas adjacent to the small areas as the independent distribution situation training samples, and the rest 15 percent of the situation small-area crop marking samples as the independent distribution situation verification samples.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a large-scale crop remote sensing classification method and system, wherein the method and system establish a migration remote sensing network RSNet model according to a residual error network model and a pyramid pooling network model, and pre-train the migration RSNet model according to historical training samples in large-scale crop marking samples in historical years to establish a pre-training migration RSNet model; the pre-training model is finely adjusted based on randomly distributed and independently distributed present small-area high-precision agricultural crop marker samples, and then the high-efficiency automatic classification of large-scale crops is realized on present images based on the fine adjustment model. The fine-tuning model for the existing small-area crop marking sample provided by the invention greatly improves the classification precision of large-scale crops on the whole, realizes the migration of the spatial scale of the crop marking sample, solves the problem that the marking sample is limited by a specific area, a specific image and a specific target in the traditional crop classification, and improves the classification precision and the classification efficiency of the large-scale crops.
In addition, the method and the system provided by the invention realize rapid and automatic classification of large-scale crops for the present images based on the pre-training migration RSNet model constructed by large-scale crop marking samples in historical years, and fully utilize the historical marking samples in the classification process, thereby greatly reducing the workload of sample selection and solving the problem of data starvation in the traditional large-scale crop classification method.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a method of the remote sensing classification method for large-scale crops according to the present invention;
FIG. 2 is a comparison graph of migration RSNet model accuracy evaluation of 3 model migration scenarios provided by the present invention;
FIG. 3 is a diagram showing the classification results of provincial crop classification using the pre-trained migration RSNet model provided by the present invention;
FIG. 4 is a diagram of the classification results of provincial crop classification using the first trimmed migration RSNet model provided by the present invention;
FIG. 5 is a diagram of the classification results of provincial crop classification using the second trimmed migration RSNet model provided by the present invention;
fig. 6 is a system structure diagram of the remote sensing classification system for large-scale crops provided by the invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the 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 invention.
The invention aims to provide a large-scale crop remote sensing classification method and system to improve classification precision and efficiency of large-scale remote sensing image classification and solve the problem of data hunger in the traditional large-scale crop remote sensing classification method.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
FIG. 1 is a flow chart of a method of the remote sensing classification method of large-scale crops provided by the invention. Referring to fig. 1, the remote sensing classification method for large-scale crops provided by the invention specifically comprises the following steps:
step 101: and obtaining a large-range crop marking sample and a present small-area crop marking sample in the historical year. The method specifically comprises the following steps:
step S1: acquiring historical multi-temporal time-sharing images and historical crop weather calendar data in a large-scale region of historical years; the large-range area is a province or above-province range area;
step S2: constructing a large-scale crop marking sample of the historical year according to the historical multi-temporal segmentation image and the historical crop weather calendar data; the method specifically comprises the following steps:
based on historical multi-temporal subdivision images (such as a high-resolution one-number 16-meter spatial resolution image) and historical crop weather data of a large range (such as in a province region) of historical years, a method of combining target change detection and a Support Vector Machine (SVM) is utilized to conduct large-scale crop remote sensing classification, and a large-range crop marking sample of the historical years is constructed.
Step S3: acquiring present multi-temporal high-resolution images of small areas in the current year and current crop climate calendar data; the small area is a plurality of counties.
Step S4: constructing a crop marking sample of the present small region according to the present multi-temporal height-score image and the current crop calendar data; the method specifically comprises the following steps:
based on the present multi-temporal high-resolution images (such as high-resolution one-number 2 m/8 m spatial resolution fusion images) in the small areas of the current year and the current crop climate data, performing image object segmentation on the present multi-temporal high-resolution images and the current crop climate data by adopting an object-oriented method to obtain segmented image objects of the designated area;
constructing spectral features and textural features of crops according to the segmented image objects in the designated area;
performing feature extraction and optimization on spectral features and textural features of the crops by using a Random Forest algorithm (RF), performing high-precision crop classification by using a support vector machine classifier, and determining a high-resolution crop classification result in a designated area;
and resampling the high-resolution crop classification result in the designated area into a resolution ratio which is the same as that of the medium-resolution crop marking sample by adopting a resampling method based on a mode principle, and obtaining the present small-area crop marking sample.
Step 102: and determining a historical training sample and a historical verification sample according to the wide-range agricultural marker sample of the historical year. The method specifically comprises the following steps:
step S3: and selecting 85% of the historical year large-range crop marking samples as the historical training samples, and taking the rest 15% of the historical year large-range crop marking samples as the historical verification samples.
Step 103: and determining randomly distributed situation training samples, randomly distributed situation verification samples, independently distributed situation training samples and independently distributed situation verification samples according to the situation small-area crop marking samples. The method specifically comprises the following steps:
and randomly selecting 85% of the crop mark samples in the present small area as the randomly distributed present training samples, and taking the rest 15% of the crop mark samples in the present small area as the randomly distributed present verification samples.
Selecting the crop marking samples of the present small areas in 85 percent of the designated areas adjacent to the small areas as the independently distributed present training samples, and using the remaining 15 percent of the crop marking samples of the present small areas as the independently distributed present verification samples.
Step 104: and acquiring a residual error network model and a pyramid pooling network model.
Step 105: and establishing a migration RSNet model according to the residual error network model and the pyramid pooling network model.
The invention improves the ResNet-50 model of the existing Residual Network (ResNet), combines the ResNet-50 Residual Network model with a Pyramid pooling Network (PSPNet) model, and generates a model suitable for multispectral Remote Sensing image crop end-to-end classification, namely the Remote Sensing Network (RSNet) model.
Step 106: and pre-training the migration RSNet model according to the historical training sample, and establishing a pre-training migration RSNet model.
And pre-training the migration RSNet model based on historical training samples in the large-scale crop marking samples in the historical years, so that the model learns and extracts the characteristics related to the remote sensing image crops, and the established pre-training migration RSNet model can be suitable for remote sensing classification of the remote sensing image large-scale crops.
Step 107: and fine-tuning the pre-training migration RSNet model according to the randomly distributed situation training samples, and establishing a first fine-tuning migration RSNet model.
And fine-tuning the pre-trained migration RSNet model which is fully pre-trained based on the randomly distributed situation training samples in the randomly distributed situation small-area crop marking samples, so that the RSNet can further learn the characteristics of the crops in the situation remote sensing image, the classification precision of the crops in the situation remote sensing image is improved, and the first fine-tuned migration RSNet model is obtained.
Step 108: and fine-tuning the pre-training migration RSNet model according to the independently distributed situation training samples, and establishing a second fine-tuning migration RSNet model.
And fine-tuning the pre-trained migration RSNet model which is fully pre-trained based on the independent distribution situation training samples in the independent distribution situation small-area crop marking samples, so that the RSNet can further learn the characteristics of the crops in the situation remote sensing image, the classification precision of the crops in the situation remote sensing image is improved, and the second fine-tuned migration RSNet model is obtained.
According to the invention, 3 migration remote sensing network model scenes are designed according to large-scale crop marking samples and present small-area crop marking samples (including randomly distributed present small-area crop marking samples and independently distributed present small-area crop marking samples) in historical years, wherein the three scenes are respectively as follows: in a first scenario, a pre-training migration RSNet model is constructed based on a large-range crop marking sample in a historical year; in a second scenario, fine tuning is carried out on the pre-training migration RSNet model based on randomly distributed small-area crop marking samples, and a first fine tuning migration RSNet model is generated; and thirdly, fine tuning the pre-training model based on the independently distributed small-area crop marking samples to generate a second fine tuning migration RSNet model. And training and precision evaluation are respectively carried out on the RSNet models of the 3 migration scenes, and indexes of the model precision evaluation are the Overall classification precision (OA) of the confusion matrix, the drawing Precision (PA) and the User precision (UA).
The training samples and the validation samples of the 3 migration RSNet models are shown in Table 1:
TABLE 1 number of training samples and validation samples for different migration RSNet models
| Scene one | Scene two | Scene three | |
| Training sample/piece | 92629 | 1335 | 1228 |
| Verify the sample/piece | 8673 | 298 | 400 |
The specific meaning and formula of each model precision evaluation index are shown in table 2:
TABLE 2 model accuracy evaluation index
The precision evaluation comparison results of the migration RSNet models (the pre-training migration RSNet model, the first fine-tuning migration RSNet model, and the second fine-tuning migration RSNet model, respectively) of the 3 model migration scenarios are shown in fig. 2. In fig. 2, CNN-1332000 is a pre-training migration RSNet model of a scenario one provided in the embodiment of the present invention, CNN-52000 is a first fine-tuning migration RSNet model of a scenario two provided in the embodiment of the present invention, CNN-24000 is a second fine-tuning migration RSNet model of a scenario three provided in the embodiment of the present invention, and a number behind CNN represents a number of model iterations in the scenario. As can be seen from fig. 2, the overall accuracy of all 3 models is greater than 80%, where the accuracy of the first fine-tuning migration RSNet model in scenario two is the highest, and the convergence speed of the second fine-tuning migration RSNet model in scenario three is the fastest.
Step 109: and carrying out provincial crop classification on the present image by adopting the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model.
Respectively classifying crops of the present images based on RSNet models (including the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model) of 3 migration scenes, and designing corresponding large-scale classification scenes, wherein the large-scale classification scenes are respectively as follows: carrying out provincial crop classification based on the pre-training migration RSNet model, and verifying the generalization of the pre-training migration RSNet model on a time scale; carrying out provincial crop classification based on the first fine-tuning migration RSNet model, and verifying the applicability and generalization capability of the first fine-tuning migration RSNet model; and thirdly, classifying provincial crops based on the second fine-tuning migration RSNet model, and verifying the generalization of the second fine-tuning migration RSNet model in space scale.
After the pre-training migration RSNet model, the first fine-tuning migration RSNet model, and the second fine-tuning migration RSNet model are adopted to classify the provincial crops of the present situation image in step 109, the method further includes:
performing precision evaluation on the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model by adopting a field investigation sample point; the indexes of the precision evaluation comprise Overall classification precision (OA) of the confusion matrix, drawing Precision (PA), user precision (UA), Kappa Coefficient (K) and F1 score representing weighted average of drawing precision and user precision.
Respectively applying the pre-training migration RSNet model (CNN-1332000), the first fine-tuning migration RSNet model (CNN-52000) and the second fine-tuning migration RSNet model (CNN-24000) to classification of provincial crops of present images, and performing precision evaluation by using field investigation sample points, wherein the precision evaluation indexes are respectively total classification precision (OA), drawing Precision (PA), user precision (UA) and kappa coefficient (formula is shown as follows)
) And F1 score characterizing charting accuracy and user accuracy weighted average, F1 ═ 2 × (PA × UA)]/[(PA+UA)]. The specific meanings and formulas of the respective accuracy evaluation indexes are shown in table 3:
TABLE 3 model accuracy evaluation index
The classification result of the provincial crop classification of the present image by adopting the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model is as follows:
(1) in the first scenario, the classification result of the provincial crop classification by using the pre-trained migration RSNet model is shown in fig. 3, and the assessment result of the classification precision of the provincial crops is shown in table 4:
TABLE 4 evaluation table for classification accuracy of provincial crops based on CNN-1332000 model
The accuracy evaluation table 4 of the classification result of the scene-pre-training model shows that the drawing accuracy of rice and corn is low, that is, a large amount of rice and corn are classified into other types. According to the fact that the difference of the topography of Jilin province is obvious, the topography inclines from the southeast to the northwest, and the obvious characteristics of high southeast and low northwest are presented, as can be seen from the graph in FIG. 3, the hilly area and the mountain area in the southeast part have obvious missing separation phenomena, and the classification result of the northwest part with flat topography is better. However, in 4 counties (area two in fig. 3) including the city of tonghui, the city of da ampere, the city of the previous guo ross mongolian city and the county of right county in northwest, the phenomenon of relatively obvious missing marks, especially missing marks of corns, is caused, the remote sensing images are analyzed to find that plots in the 4 counties are relatively broken and the image quality is poor, wherein the southern part of the city of da ampere is covered by thick clouds, the images in the bare places of the city of tonghui and the county of right county are covered by thin clouds, the city of tonghui, the city of the previous guo ross mongolian city and the county of right county have a large amount of saline-alkali soil distribution, and the images are white, so that the spectral characteristics of the corns are not obvious. In the southeast region with large area fluctuation, the plot is also broken, and the model predicts that the small class of the broken region is convoluted into the large class in the convolution process, so that the rice and the corn which are distributed sporadically are wrongly classified into other classes. However, a better classification effect is achieved in a region (region one in fig. 3) with flat terrain, regular land parcels and continuous crop planting. Particularly, the scene-model obtains good classification results in elm markets with flat topography and regular plots in the middle and north, the overall classification precision reaches 90.88%, the Kappa coefficient is 0.86, the drawing precision and the user precision of rice are both above 90.27%, and the drawing precision and the user precision of corn are both above 85.54%. More particularly, as the remote sensing image of the elm market in 2016 has a large amount of thick cloud coverage, no crop classification is performed, that is, the elm market does not participate in model pre-training or model fine-tuning, but the crop classification result predicted by the pre-training migration RSNet model constructed based on historical large-range crop marking samples is good. The experimental result shows that the pre-training migration RSNet model provided by the invention has better time scale generalization in the region with flat terrain and regular plots, solves the problem of data hunger of a large number of marked samples in large-scale remote sensing data classification to a certain extent, and realizes the rapid automatic remote sensing classification of large-scale crops.
(2) The classification result of the second scenario, which uses the first fine-tuning migration RSNet model to classify the provincial crops, is shown in fig. 4, and the evaluation result of the classification precision of the provincial crops is shown in table 5:
TABLE 5 provincial crop classification accuracy evaluation table based on CNN-52000 model
The classification result of the second fine-tuning migration RSNet model for provincial crop classification in the third scenario is shown in fig. 5, and the evaluation result of the provincial crop classification accuracy is shown in table 6:
TABLE 6 provincial crop classification accuracy evaluation table based on CNN-24000 model
On the whole, the classification precision of rice and corn in the second scene and the third scene is greatly improved, and two fine tuning models (the first fine tuning migration RSNet model and the second fine tuning migration RSNet model) obtain consistent classification results.
The accuracy evaluation tables 5 and 6 show that the drawing accuracy of rice is low, that is, a large amount of rice is missed. As can be seen from fig. 4 and 5, compared with the classification result of the scene-pre-trained migration RSNet model, the southeast region with large topography is improved greatly (region three in fig. 4 and 5), but the miss-distinguishing phenomenon still exists. In the middle region where the terrain is flat (region one in fig. 4 and 5), the classification results of both fine-tuned models are better. However, severe scoring phenomena still exist in the northwest region (region two in fig. 4 and 5), wherein the 4 prefectures of the city of garden, da ampere city, the city of the national county of mongolia of guo of guo ruoshi, and the county of yurt have the same phenomenon as the result of the scene-one model. Although the terrain is flat, the rice misclassification phenomenon is serious, the land parcel in the region is also broken through the analysis of classification results and remote sensing images, and more importantly, the rice spectrum difference is obvious and is greatly different from the rice spectrum in small region samples participating in fine adjustment, so that the rice characteristics extracted by a fine adjustment model are not comprehensive, and the rice classification precision is not obviously improved. Taking an elm city as an example, the CNN-1332000 model fully pre-trained by using a historical training sample obtains good effect in the classification of the elm city, and the CNN-52000 and CNN-24000 models finely adjusted based on the current small region training sample have the overall classification precision of 79.52 percent in the classification of the elm city, which is lower than the classification result of the CNN-1332000 model; but the user precision and drawing precision of the corn are both more than 90 percent and higher than the classification result of the CNN-1332000 model; the user precision of rice reaches 96.03%, and the drawing precision is only 61.88% due to large rice spectrum difference, which is lower than the classification result of the CNN-1332000 model. This is because, compared to corn, rice belongs to a subclass, the rice area of the present marker sample is much smaller than that of corn and others, and the spectral feature difference of rice in the whole province is obvious, so that the whole precision of rice is not obviously improved, and fully trained corn with large area ratio has good classification effect and obviously improved precision. The classification results of the scene two and the scene three verify the application capability of the fine-tuning migration RSNet model in a large-scale range, solve the problem that a training sample in a traditional classification model is only suitable for a specific region, a specific image and a specific target, reduce the workload of sample selection and improve the classification efficiency.
In summary, from the 3 classification scenarios described above, it is found that: firstly, the method classifies crops in 60 counties of Jilin province based on the migration RSNet model, and the time is only 6.5 hours, so that the classification speed is greatly improved. Secondly, the result of directly predicting the crops in the present image by using the historical model shows that better classification results are obtained in areas with flat terrain, regular plots and continuous crop planting, and the generalization capability in broken plots is to be improved. On the whole, the migration RSNet model fully utilizes the historical training samples, so that the migration of time scale and space scale is realized, the problem that a large number of marked samples are needed for supporting in remote sensing image crop classification is solved to a certain extent, and the automatic and rapid remote sensing image crop classification is realized. Finally, the result of carrying out crop classification on the terrain remote sensing image by utilizing the small-area terrain marking sample fine-tuning pre-model shows that corns which are wide in planting distribution, large in area and large in scale are fully trained in a flat plain area with flat terrain and hills and mountain areas with large terrain fluctuation, good classification results are obtained, the improvement of the classification accuracy of rice with small planting area and large spectrum difference is not obvious, and the problems of unbalanced samples and sample representativeness are to be improved. On the whole, the RSNet model is finely adjusted and pre-trained on the basis of the small-area situation marking samples, the defect that the training samples in the traditional classification model are suitable for specific areas, specific images and specific targets is overcome, the workload of sample selection is reduced, the classification efficiency is improved, and a solid foundation is laid for forming a set of large-scale crop remote sensing classification method.
Fig. 6 is a system structure diagram of the remote sensing classification system for large-scale crops provided by the invention. Referring to fig. 6, the remote sensing classification system for large-scale crops provided by the invention comprises:
the crop marking sample acquisition module 601 is used for acquiring large-scale crop marking samples and present small-area crop marking samples in historical years;
a historical sample determining module 602, configured to determine a historical training sample and a historical verification sample according to the wide-range agricultural marker sample of the historical year;
a situation sample determination module 603, configured to determine, according to the situation small-area crop marker sample, a randomly-distributed situation training sample, a randomly-distributed situation verification sample, an independently-distributed situation training sample, and an independently-distributed situation verification sample;
a network model obtaining module 604, configured to obtain a residual network model and a pyramid pooling network model;
a migration RSNet model establishing module 605, configured to establish a migration RSNet model according to the residual network model and the pyramid pooling network model;
a pre-training migration RSNet model establishing module 606, configured to pre-train the migration RSNet model according to the historical training samples, and establish a pre-training migration RSNet model;
a first fine-tuning migration RSNet model establishing module 607, configured to perform fine tuning on the pre-training migration RSNet model according to the randomly distributed situation training samples, and establish a first fine-tuning migration RSNet model;
a second fine-tuning migration RSNet model 608, configured to perform fine tuning on the pre-training migration RSNet model according to the independently distributed presence training samples, and establish a second fine-tuning migration RSNet model;
a crop classification module 609, configured to perform provincial crop classification on the present image by using the pre-training migration RSNet model, the first fine-tuning migration RSNet model, and the second fine-tuning migration RSNet model.
The agricultural marker sample acquisition module 601 specifically includes:
the historical image and historical data acquisition unit is used for acquiring historical multi-temporal segmentation images and historical crop calendar data in a large-scale area of a historical year; the large-range area is a province or above-province range area;
the historical year large-scale crop marking sample construction unit is used for constructing the historical year large-scale crop marking sample according to the historical multi-time-phase separated image and the historical crop weather calendar data;
the current image and current data acquisition unit is used for acquiring the current multi-temporal high-resolution images of the current small region of the year and the current crop calendar data; the small areas are a plurality of counties;
and the present small-area crop marking sample construction unit is used for constructing the present small-area crop marking sample according to the present multi-temporal high-resolution image and the current crop climate data.
The construction unit of the large-scale crop marking sample in the historical year specifically comprises the following steps:
and the historical year large-range crop marking sample constructing subunit is used for classifying crops by adopting a method of combining target change detection and a support vector machine according to the historical multi-time-phase time-division image and the historical crop weather calendar data, and constructing the historical year large-range crop marking sample.
The construction unit of the crop marking sample in the small area of the present situation specifically comprises:
the image segmentation subunit is used for carrying out image object segmentation on the current multi-temporal high-resolution image and the current crop calendar data by adopting an object-oriented method to obtain a segmented designated area image object;
the characteristic construction subunit is used for constructing spectral characteristics and textural characteristics of crops according to the segmented image objects in the designated area;
the high-precision crop classification subunit is used for extracting and optimizing the spectral features and the textural features of the crops by adopting a random forest algorithm, classifying the high-precision crops by using a support vector machine classifier and determining a high-resolution crop classification result;
and the resampling subunit is used for resampling the high-resolution crop classification result into a resolution which is the same as that of the medium-resolution crop marking sample by adopting a resampling method based on a mode principle, and obtaining the present small-area crop marking sample.
The history sample determining module 602 specifically includes:
and the historical sample determining unit is used for selecting 85% of the historical year large-range crop marking samples as the historical training samples, and selecting the rest 15% of the historical year large-range crop marking samples as the historical verification samples.
The present sample determination module 603 specifically includes:
a randomly distributed situation sample determination unit, configured to randomly select 85% of the situation small-area crop marker samples as the randomly distributed situation training samples, and the remaining 15% of the situation small-area crop marker samples as the randomly distributed situation verification samples;
and the independent distribution situation sample determination unit is used for selecting the situation small-area crop marking samples in 85 percent of the designated areas adjacent to the small areas as the independent distribution situation training samples, and the rest 15 percent of the situation small-area crop marking samples as the independent distribution situation verification samples.
In addition, the system further comprises an accuracy evaluation module for performing accuracy evaluation on the pre-training migration RSNet model, the first fine-tuning migration RSNet model and the second fine-tuning migration RSNet model by adopting a field investigation sample point; the indexes of the precision evaluation comprise overall classification precision, drawing precision, user precision, Kappa coefficient and F1 score representing weighted average of the drawing precision and the user precision.
The large-scale crop remote sensing classification system performs model pre-training on the migration RSNet model based on the historical year large-scale crop marking samples, and performs fine adjustment on the pre-trained migration RSNet model based on the randomly distributed present small-area crop marking samples and the independently distributed present small-area crop marking samples, so that the remote sensing rapid automatic classification of the present image provincial crops is realized, and the problem of data starvation of the training samples in the remote sensing image classification is solved.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A remote sensing classification method for large-scale crops is characterized by comprising the following steps:
obtaining a large-range crop marking sample and a present small-area crop marking sample in a historical year;
determining a historical training sample and a historical verification sample according to the wide-range agricultural marker samples of the historical years;
determining randomly distributed situation training samples, randomly distributed situation verification samples, independently distributed situation training samples and independently distributed situation verification samples according to the situation small-area crop marking samples;
acquiring a residual error network model and a pyramid pooling network model;
establishing a migration remote sensing network RSNet model according to the residual error network model and the pyramid pooling network model;
pre-training the migration remote sensing network RSNet model according to the historical training sample, and establishing a pre-training migration remote sensing network RSNet model;
fine tuning the pre-training migration remote sensing network RSNet model according to the randomly distributed situation training samples, and establishing a first fine tuning migration remote sensing network RSNet model;
fine tuning the pre-training migration remote sensing network RSNet model according to the independently distributed situation training samples, and establishing a second fine tuning migration remote sensing network RSNet model;
carrying out provincial crop classification on the present image by adopting the pre-training migration remote sensing network RSNet model, the first fine-tuning migration remote sensing network RSNet model and the second fine-tuning migration remote sensing network RSNet model;
the method for acquiring the large-range crop marking samples and the small-area crop marking samples in the historical years specifically comprises the following steps:
acquiring historical multi-temporal time-sharing images and historical crop weather calendar data in a large-scale region of historical years; the large-range area is a province or above-province range area;
constructing a large-scale crop marking sample of the historical year according to the historical multi-temporal segmentation image and the historical crop weather calendar data;
acquiring present multi-temporal high-resolution images of small areas in the current year and current crop climate calendar data; the small areas are a plurality of counties;
constructing a crop marking sample of the present small region according to the present multi-temporal height-score image and the current crop calendar data;
the construction of the crop marking sample of the small region of the present situation according to the present multi-temporal high-resolution image and the current crop calendar data specifically comprises:
performing image object segmentation on the present multi-temporal high-resolution image and the current crop climate data by adopting an object-oriented method to obtain a segmented image object of a designated area;
constructing spectral features and textural features of crops according to the segmented image objects in the designated area;
performing feature extraction and optimization on the spectral features and the textural features of the crops by adopting a random forest algorithm, performing high-precision crop classification by using a support vector machine classifier, and determining a high-resolution crop classification result;
and resampling the high-resolution crop classification result into a resolution which is the same as that of the medium-resolution crop marking sample by adopting a resampling method of a mode principle, and obtaining the present small-area crop marking sample.
2. The remote sensing classification method for large-scale crops according to claim 1, wherein the constructing of the large-scale crop marking sample in the historical year according to the historical multi-temporal segmentation image and the historical crop climate history data specifically comprises:
and classifying the crops by adopting a method of combining target change detection and a support vector machine according to the historical multi-temporal segmentation images and the historical crop calendar data, and constructing the large-range crop marking samples of the historical years.
3. The remote sensing classification method for large-scale crops according to claim 1, wherein the determining of historical training samples and historical verification samples according to the large-scale agricultural marker samples in the historical years specifically comprises:
and selecting 85% of the historical year large-range crop marking samples as the historical training samples, and taking the rest 15% of the historical year large-range crop marking samples as the historical verification samples.
4. The remote sensing classification method for large-scale crops according to claim 1, wherein the determining of the randomly distributed situation training samples, the randomly distributed situation verification samples, the independently distributed situation training samples and the independently distributed situation verification samples according to the situation small-area crop marking samples specifically comprises:
randomly selecting 85% of the crop marking samples in the present small area as the randomly distributed present training samples, and taking the rest 15% of the crop marking samples in the present small area as the randomly distributed present verification samples;
selecting the crop marking samples of the present small areas in 85 percent of the designated areas adjacent to the small areas as the independently distributed present training samples, and using the remaining 15 percent of the crop marking samples of the present small areas as the independently distributed present verification samples.
5. The remote sensing classification method for large-scale crops according to claim 1, wherein after the pre-trained migration remote sensing network RSNet model, the first fine-tuning migration remote sensing network RSNet model and the second fine-tuning migration remote sensing network RSNet model are adopted to classify the present images into provincial crops, the method further comprises the following steps:
carrying out precision evaluation on the pre-training migration remote sensing network RSNet model, the first fine-tuning migration remote sensing network RSNet model and the second fine-tuning migration remote sensing network RSNet model by adopting a field investigation sample point; the indexes of the precision evaluation comprise overall classification precision, drawing precision, user precision, Kappa coefficient and F1 score representing weighted average of the drawing precision and the user precision.
6. A remote sensing classification system for large-scale crops is characterized by comprising:
the crop marking sample acquisition module is used for acquiring large-range crop marking samples and present small-area crop marking samples in historical years;
obtaining a crop marking sample of the present small area, specifically comprising:
acquiring present multi-temporal high-resolution images of small areas in the current year and current crop climate calendar data; the small areas are a plurality of counties;
constructing a crop marking sample of the present small region according to the present multi-temporal height-score image and the current crop calendar data;
the construction of the crop marking sample of the small region of the present situation according to the present multi-temporal high-resolution image and the current crop calendar data specifically comprises:
performing image object segmentation on the present multi-temporal high-resolution image and the current crop climate data by adopting an object-oriented method to obtain a segmented image object of a designated area;
constructing spectral features and textural features of crops according to the segmented image objects in the designated area;
performing feature extraction and optimization on the spectral features and the textural features of the crops by adopting a random forest algorithm, performing high-precision crop classification by using a support vector machine classifier, and determining a high-resolution crop classification result;
resampling the high-resolution crop classification result into a resolution which is the same as that of the medium-resolution crop marking sample by adopting a resampling method of a mode principle to obtain the present small-area crop marking sample;
the historical sample determining module is used for determining a historical training sample and a historical verification sample according to the wide-range agricultural marker samples of the historical years;
the situation sample determination module is used for determining randomly distributed situation training samples, randomly distributed situation verification samples, independently distributed situation training samples and independently distributed situation verification samples according to the situation small-area crop marking samples;
the network model acquisition module is used for acquiring a residual network model and a pyramid pooling network model;
the migration remote sensing network RSNet model establishing module is used for establishing a migration remote sensing network RSNet model according to the residual error network model and the pyramid pooling network model;
the pre-training migration remote sensing network RSNet model establishing module is used for pre-training the migration remote sensing network RSNet model according to the historical training sample and establishing a pre-training migration remote sensing network RSNet model;
the first fine-tuning migration remote sensing network RSNet model building module is used for fine-tuning the pre-training migration remote sensing network RSNet model according to the randomly distributed situation training samples and building a first fine-tuning migration remote sensing network RSNet model;
the second fine tuning migration remote sensing network RSNet model is used for fine tuning the pre-training migration remote sensing network RSNet model according to the independent distribution situation training sample and establishing a second fine tuning migration remote sensing network RSNet model;
and the crop classification module is used for classifying the provincial crops of the present images by adopting the pre-training migration remote sensing network RSNet model, the first fine-tuning migration remote sensing network RSNet model and the second fine-tuning migration remote sensing network RSNet model.
7. The remote sensing classification system for large-scale crops according to claim 6, wherein the historical sample determination module specifically comprises:
and the historical sample determining unit is used for selecting 85% of the historical year large-range crop marking samples as the historical training samples, and selecting the rest 15% of the historical year large-range crop marking samples as the historical verification samples.
8. The remote sensing classification system for large-scale crops as claimed in claim 6, wherein the situation sample determination module specifically comprises:
a randomly distributed situation sample determination unit, configured to randomly select 85% of the situation small-area crop marker samples as the randomly distributed situation training samples, and the remaining 15% of the situation small-area crop marker samples as the randomly distributed situation verification samples;
and the independent distribution situation sample determination unit is used for selecting the situation small-area crop marking samples in 85 percent of the designated areas adjacent to the small areas as the independent distribution situation training samples, and the rest 15 percent of the situation small-area crop marking samples as the independent distribution situation verification samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811183899.XA CN109145885B (en) | 2018-10-11 | 2018-10-11 | Remote sensing classification method and system for large-scale crops |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811183899.XA CN109145885B (en) | 2018-10-11 | 2018-10-11 | Remote sensing classification method and system for large-scale crops |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109145885A CN109145885A (en) | 2019-01-04 |
| CN109145885B true CN109145885B (en) | 2020-09-15 |
Family
ID=64811052
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811183899.XA Active CN109145885B (en) | 2018-10-11 | 2018-10-11 | Remote sensing classification method and system for large-scale crops |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145885B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110298366B (en) * | 2019-07-05 | 2021-05-04 | 北华航天工业学院 | Crop distribution extraction method and device |
| CN112840348B (en) * | 2019-10-11 | 2024-05-03 | 安徽中科智能感知科技股份有限公司 | Crop planting distribution prediction method based on time sequence remote sensing data and convolutional neural network |
| CN113033453B (en) * | 2021-04-06 | 2023-09-15 | 北京艾尔思时代科技有限公司 | Method and system suitable for remote sensing identification of crop types in landscape crushing area |
| CN114332570B (en) * | 2022-03-17 | 2022-06-03 | 北京艾尔思时代科技有限公司 | Method and system for carrying out crop migration classification based on CDL deep learning |
| CN117253155B (en) * | 2023-11-17 | 2024-03-15 | 山东大学 | Human activity detection method and system based on deep learning |
| CN117975292B (en) * | 2024-03-25 | 2024-06-14 | 中国科学院东北地理与农业生态研究所 | Extraction method for temperate forest swamps |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053408A (en) * | 2017-12-05 | 2018-05-18 | 中国资源卫星应用中心 | A kind of high automation updating investigation of land use based on satellite-remote-sensing image |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9152881B2 (en) * | 2012-09-13 | 2015-10-06 | Los Alamos National Security, Llc | Image fusion using sparse overcomplete feature dictionaries |
| CN103761447B (en) * | 2014-02-18 | 2017-05-24 | 中国科学院遥感与数字地球研究所 | Planting area remote sensing confirming method for grain subsidy accounting of prefecture level and county level |
-
2018
- 2018-10-11 CN CN201811183899.XA patent/CN109145885B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053408A (en) * | 2017-12-05 | 2018-05-18 | 中国资源卫星应用中心 | A kind of high automation updating investigation of land use based on satellite-remote-sensing image |
Non-Patent Citations (1)
| Title |
|---|
| Fully Convolutional Neural Networks for Large Scale Cropland Mapping with Historical Label Dataset;Dujuan Zhang等;《IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium》;20180727;第4659-4662页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109145885A (en) | 2019-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109145885B (en) | Remote sensing classification method and system for large-scale crops | |
| CN109409261B (en) | Crop classification method and system | |
| CN106951836B (en) | crop coverage extraction method based on prior threshold optimization convolutional neural network | |
| CN105740759B (en) | Semilate rice information decision tree classification approach based on feature extraction in multi-temporal data | |
| CN110796168A (en) | Improved YOLOv 3-based vehicle detection method | |
| CN108984803B (en) | Method and system for spatializing crop yield | |
| CN111310756B (en) | Damaged corn particle detection and classification method based on deep learning | |
| CN108846832A (en) | A kind of change detecting method and system based on multi-temporal remote sensing image and GIS data | |
| Liu et al. | Establishing a citywide street tree inventory with street view images and computer vision techniques | |
| CN105792353A (en) | Image matching type indoor positioning method with assistance of crowd sensing WiFi signal fingerprint | |
| CN112749627A (en) | Method and device for dynamically monitoring tobacco based on multi-source remote sensing image | |
| CN110263717A (en) | It is a kind of incorporate streetscape image land used status determine method | |
| CN111340826A (en) | Single tree crown segmentation algorithm for aerial image based on superpixels and topological features | |
| CN104156734A (en) | Fully-autonomous on-line study method based on random fern classifier | |
| CN111241939A (en) | Rice yield estimation method based on unmanned aerial vehicle digital image | |
| CN111598045A (en) | Remote sensing farmland change detection method based on object spectrum and mixed spectrum | |
| CN104063713A (en) | Semi-autonomous on-line studying method based on random fern classifier | |
| CN110826556A (en) | Broad-spectrum crop weed identification and positioning method for improving deep learning | |
| CN113205014B (en) | Time sequence data farmland extraction method based on image sharpening | |
| CN112084926A (en) | Screening method for ecological restoration plants of abandoned mine | |
| CN116129260A (en) | Forage grass image recognition method based on deep learning | |
| Zhang et al. | Enhancing model performance in detecting lodging areas in wheat fields using UAV RGB Imagery: Considering spatial and temporal variations | |
| CN113469122A (en) | Deep learning based crop space-time generalization classification method and system | |
| CN117197595A (en) | Fruit tree growth period identification method, device and management platform based on edge calculation | |
| CN109726679B (en) | Remote sensing classification error spatial distribution mapping method |
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 |