Multi-source remote sensing data space sample set construction method, information extraction method and system
By using a random classification and multi-source remote sensing image data extraction system, the construction of remote sensing data sample sets and model training process is simplified, solving the problems of complexity, cumbersomeness and high threshold in traditional methods, and realizing efficient and intuitive remote sensing information extraction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2026-04-10
AI Technical Summary
The traditional process of constructing remote sensing data sample sets is complex, time-consuming, and labor-intensive. It is also subject to human subjective errors, which affect the objectivity of the sample set and the performance of the model. The difficulty in adjusting the model training parameters increases the technical threshold, resulting in low efficiency of machine learning in remote sensing information extraction.
The training and test sample regions are divided using a random classification method. Combined with expert experience, the sample set production process is simplified. The model is trained and its accuracy is verified through a multi-source remote sensing image data extraction system. An intuitive system interface is provided to reduce the difficulty of manual operation.
It improves the objectivity of the sample set and the generalization ability of the model, reduces the intensity of manual operation, simplifies the machine learning process, lowers the technical threshold, improves efficiency, and adapts to large-scale data processing.
Smart Images
Figure CN121837812A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of remote sensing image processing, and particularly relates to a multi-source remote sensing data space sample set construction method, an information extraction method and a system. BACKGROUND
[0002] In recent years, with the rapid increase in the number of satellites, the scale of remote sensing data has also shown an explosive growth trend. On the one hand, the growth in the number of satellites has accelerated the monitoring frequency of the earth's surface and improved the spatial resolution and timeliness of the data. On the other hand, the massive amount of remote sensing data has brought great challenges to remote sensing industry workers. How to effectively process and analyze these massive amounts of data to extract useful information and provide reference and decision-making basis for solving problems in the fields of environment, climate, agriculture, urban planning, etc. is particularly important for the processing and analysis of remote sensing data.
[0003] Machine learning algorithms have proven their strong capabilities in feature recognition and classification, regression. These algorithms can use complex networks and structures to capture the features of input data and establish explicit relationships between output variables. Machine learning models can learn from large amounts of data and perform pattern recognition, constantly improving their performance and accuracy. The development of this capability has enabled machine learning to play an important role in various fields. Machine learning technology is widely used in the field of remote sensing. Machine learning algorithms can effectively analyze, process and interpret remote sensing data. Using machine learning technology, remote sensing images can be used for target detection, identification of land cover types, monitoring of natural disasters, tracking of environmental changes, etc.
[0004] The overall process of traditional machine learning technology for remote sensing data information extraction involves sample set production, model training and precision verification, etc. However, the above steps have the following technical defects:
[0005] (1) Sample set production is the basis for machine learning remote sensing information extraction, and is characterized by complex and tedious operation, time-consuming and labor-intensive, and is crucial to the performance of machine learning models. Due to the complexity and diversity of massive amounts of remote sensing data, traditional sample set construction often requires manual annotation and manual allocation to divide the complex spatial sample area into training sample area and test sample area. The above process not only consumes a large amount of time and human resources, but also largely accumulates individual subjective experience and judgment errors, thereby affecting the objectivity of the sample set. Especially when the data set has spatial or temporal correlation, the randomness of the test set and the validation set may lead to over-optimization of the model for a particular subset, resulting in poor performance in actual application. Therefore, the machine learning sample set production problem for remote sensing information extraction urgently needs a more intelligent and efficient method to reduce the burden on manual labor and improve the objectivity of the sample set, so as to more effectively cope with the challenges of large sample sets.
[0006] (2) Model training is one of the core links of machine learning remote sensing information extraction. When constructing a machine learning model for remote sensing information extraction, the corresponding parameters need to be continuously optimized according to the samples in the model training process. Generally, when using machine learning methods (such as random forest, support vector machine, multilayer perceptron, etc.) to carry out remote sensing image classification, the parameters of the model need to be adjusted, such as the number of decision trees forming the random forest, the kernel function and the penalty coefficient of the support vector machine, in order to achieve the best effect. However, the above model parameter adjustment process requires a strong code basis, which increases the difficulty of remote sensing information extraction method, and makes the use of machine learning method for remote sensing information extraction have certain technical threshold. SUMMARY
[0007] The purpose of the present application is to overcome the shortcomings of the prior art, and provide a multi-source remote sensing data spatial sample set construction method, information extraction method and system.
[0008] The present application specifically introduces random classification on the basis of the samples selected by human beings, and respectively forms a training sample area and a test sample area by random allocation. This method combines the advantages of expert experience and random algorithm, improves the quality of the sample set, and greatly reduces the intensity of human work. Moreover, through the established multi-source remote sensing image data extraction system, a large number of machine learning samples can be used for model training and precision verification to achieve random sampling and feature extraction of multi-source remote sensing data. This method simplifies the entire process of remote sensing information extraction using machine learning method, avoids the code debugging process of remote sensing workers, and makes parameter debugging more intuitive and efficient through an intuitive system interface.
[0009] Based on the above technical idea, the first object of the present application is to provide a multi-source remote sensing data spatial sample set construction method, comprising the following steps:
[0010] Step 1, pre-processing the multi-source remote sensing image, and creating a spatial total sample area;
[0011] Step 2, randomly dividing the total sample area obtained in step 1 into a spatial training sample area and a spatial test sample area;
[0012] Step 3, based on multi-source remote sensing data, extracting attribute features from the spatial training sample area and the spatial test sample area obtained in step 2, respectively, to obtain a training sample set and a test sample set.
[0013] Preferably, step 1 comprises the following steps:
[0014] Step 11, pre-processing the multi-source remote sensing image to obtain a pre-processed image;
[0015] Step 12, selecting a target region on the pre-processed image obtained in step 11, and creating spatial sample regions of different categories on the target region;
[0016] Step 13, respectively performing label assignment on the spatial sample regions of different categories obtained in step 12, to obtain a spatial total sample region with label values.
[0017] Preferably, in step 12, the categories of the spatial sample regions include an interest region and a non-interest region, and in step 13, the interest region is respectively assigned labels according to categories, and the label of the non-interest region is 0.
[0018] Preferably, step 2 comprises the following steps:
[0019] According to the spatial total sample region obtained in step 1, the spatial total sample region obtained in step 1 is randomly divided into a spatial training sample region and a spatial test sample region according to a set division ratio.
[0020] Preferably, step 3 comprises the following steps:
[0021] Based on the multi-source remote sensing data, the feature values at each label in the spatial training sample region and the spatial test sample region obtained in step 2 are extracted respectively, to obtain a training sample set and a test sample set.
[0022] The second object of the present application is to provide a training sample set and a test sample set constructed by the multi-source remote sensing data spatial sample set construction method.
[0023] The third object of the present application is to provide a multi-source remote sensing data information extraction method, which is realized by using the training sample set and the test sample set, and comprises the following steps:
[0024] S1, selecting a machine learning model according to requirements to obtain a target machine learning model, setting model parameters, inputting the training sample set into the target machine learning model for model training, and obtaining a trained model;
[0025] S2, performing model accuracy verification on the trained model obtained in step S1 by using the test sample set, adjusting the parameters of the trained model according to the verification result, and obtaining an optimized model;
[0026] S3, inputting target remote sensing image data into the optimized model obtained in step S2 for model prediction, and obtaining a remote sensing image information extraction result.
[0027] Preferably, in step S1, the machine learning model is any one of random forest, support vector machine, decision tree, multilayer perceptron, logistic, partial least squares, xgboost, lightbm, catboost and adaboost.
[0028] Preferably, step S2 comprises the following steps:
[0029] S21, input the test sample set into the trained model obtained in step S1 to obtain a test sample prediction result, and the prediction result of the test sample is verified by using a confusion matrix to obtain a precision verification result, and the formula of the precision verification is:
[0030]
[0031] Wherein, OA represents the overall accuracy, TP represents the true class, TN represents the true negative class, FP represents the false positive class, and FN represents the false negative class; p o is the observed consistency, i.e. the overall accuracy, p e is the expected consistency, i.e. the accuracy of random classification;
[0032] S22, according to the precision verification result obtained in step S21, the parameters of the trained model are adjusted to obtain an optimized model, specifically:
[0033] When the precision verification result OA and the Kappa coefficient obtained in step S21 are both greater than 0.8, it is determined that the trained model meets the requirements, and the optimized model is directly obtained;
[0034] When the precision verification result OA and the Kappa coefficient obtained in step S21 are both not greater than 0.8, it is determined that the trained model does not meet the requirements, the parameters of the model are adjusted, and then steps S1 and S2 are repeated until the precision verification result OA and the Kappa coefficient obtained in step S21 are both greater than 0.8, it is determined that the trained model meets the requirements, and the optimized model is obtained.
[0035] The fourth object of the application is to provide a multi-source remote sensing data information extraction system, which is realized by the above extraction method, comprising:
[0036] A sample area creation module is configured to pre-process the multi-source remote sensing image and create a total sample area in space.
[0037] A sample area division module is configured to randomly divide the total sample area into a spatial training sample area and a spatial test sample area.
[0038] An attribute feature extraction module is configured to extract attribute features from the spatial training sample area and the spatial test sample area respectively to obtain a training sample set and a test sample set.
[0039] a model training module, configured to select a machine learning model according to a requirement, obtain a target machine learning model, set a model parameter, input a training sample set into the target machine learning model for model training, and obtain a trained model;
[0040] a model verification module, configured to perform model precision verification on the trained model by using a test sample set, adjust a parameter of the trained model according to a verification result, and obtain an optimized model;
[0041] a data extraction module, configured to input target remote sensing image data into the optimized model for model prediction, and obtain a remote sensing image data extraction result.
[0042] Compared with the prior art, the present application has the following beneficial effects:
[0043] (1) The multi-source remote sensing image data extraction method provided by the present application adopts a machine learning method, and training sample regions and test sample regions are respectively formed in an existing total sample region by a random allocation manner, so that the labor intensity of manually dividing training samples and test samples is effectively reduced.
[0044] (2) The multi-source remote sensing image data extraction system provided by the present application can be used for mass production of machine learning samples for model training and precision verification, so as to achieve spatial sample random sampling and feature extraction of multi-source remote sensing data. The system simplifies the whole process of remote sensing information extraction by using a machine learning method, avoids the code debugging process of remote sensing workers, and makes parameter debugging more intuitive and efficient through an intuitive system interface. In addition, the system also provides the functions of model training and precision verification, so that users can select a training model and adjust the parameters of the model through the system, thereby realizing model training and precision verification more conveniently and efficiently.
[0045] (3) Through the multi-source remote sensing image data extraction system provided by the present application, users do not need to deeply understand machine learning programming methods, but only need to complete the whole process of sample set production, model training and precision verification through simple operations, thereby greatly reducing the technical threshold, improving the efficiency, enabling more people to participate in remote sensing image processing and analysis work, and promoting the application and development of remote sensing technology in various fields.
[0046] (4) The present application can make the sample space distribution of the sample set contacted by the model in the training more uniform, and the number distribution of samples of each category more reasonable, so as to better adapt to the overall data distribution, rather than excessively focusing on some specific features of the training set, which helps to improve the generalization ability of the model, better adapt to new unknown data, and help to reduce the risk of overfitting. BRIEF DESCRIPTION OF DRAWINGS
[0047] Figure 1 The flowchart of the machine learning-based multi-source remote sensing image data extraction method provided for embodiment 1 of the present application is shown in the figure.
[0048] Figure 2 The schematic diagram of step S2 in embodiment 1 of the present application is shown in the figure.
[0049] Figure 3 The schematic diagram of step S3 in embodiment 1 of the present application is shown in the figure.
[0050] Figure 4 The interface diagram of the machine learning-based multi-source remote sensing image data extraction system provided for embodiment 2 of the present application is shown in the figure.
[0051] Figure 5 The operation interface diagram of the sample area division module in embodiment 2 of the present application is shown in the figure.
[0052] Figure 6 The operation interface diagram of the attribute feature extraction module in embodiment 2 of the present application is shown in the figure.
[0053] Figure 7 The specific meaning of the sample set (CSV format file) extracted by the attribute feature extraction module in embodiment 2 of the present application is shown in the figure.
[0054] Figure 8 The operation interface diagram of the model training module in embodiment 2 of the present application is shown in the figure.
[0055] Figure 9 The operation interface diagram of the model verification module in embodiment 2 of the present application is shown in the figure. DETAILED DESCRIPTION
[0056] The technical solutions in the embodiments of the present application will be described below in conjunction with the accompanying drawings. Figures 1 to 9 The technical solutions in the embodiments of the present application will be described below in conjunction with the accompanying drawings.
[0057] Embodiment 1
[0058] AsFigure 1 As shown, the embodiment of the present application provides a multi-source remote sensing data space sample set construction method, which specifically comprises the following steps:
[0059] S1, pre-processing the multi-source remote sensing image, and creating a total sample area, specifically comprising the following steps:
[0060] S11, pre-processing the multi-source remote sensing image to obtain a pre-processed image; the image pre-processing in the embodiment of the present application includes radiation calibration, atmospheric correction, geometric correction and geographic registration, the radiation calibration is used to eliminate the error of the sensor itself, and determine the accurate radiation value at the entrance of the sensor; the atmospheric correction is used to eliminate the atmospheric influence; the geometric correction is used to eliminate the geometric distortion and ensure the geographic accuracy; the geographic registration is to match two images from the geographic coordinates, which is convenient for spatial analysis. And each of the above pre-processing methods is a existing multi-source remote sensing data pre-processing method, so it will not be repeated here;
[0061] S12, using the labeling function of the image labeling software to select a target region (roi) on the pre-processed image obtained in step S11, and create different categories of space sample regions on the target region; the image labeling software in the embodiment of the present application is an existing image labeling software, which can realize the remote sensing data labeling function, and is preferably ENVI software.
[0062] S13, respectively labeling the different categories of space sample regions obtained in step 12 to obtain a space total sample region with label values. In the embodiment of the present application, the categories of the space sample region include an interest region and a non-interest region, and in step S13, the interest region is respectively labeled according to the category; the label value of the non-interest region is 0.
[0063] S2, randomly dividing the space total sample region obtained in step S1 into a space training sample region and a space test sample region, as shown in Figure 2 The specific steps are as follows:
[0064] S21, based on the total sample region obtained in step S1, obtaining geographic location information of the multi-source remote sensing image; in the embodiment of the present application, specifically: for the total sample region with labels obtained in step S1, obtaining the geographic location information of the multi-source remote sensing image based on Python language gdal, the geographic location information describes the position and scale of the image in the geographic space;
[0065] S22, according to the geographic location information obtained in step 21, inputting a division ratio autonomously according to different requirements, and dividing the space total sample region obtained in step 1 into a space training sample region and a space test sample region according to the set division ratio.
[0066] S3, based on multi-source remote sensing data, attribute features are extracted from the training sample region and the test sample region obtained in step S2 respectively, and a training sample set and a test sample set are obtained, as shown in the following table: Figure 3 As shown in the following table:
[0067] Based on multi-source remote sensing data, the feature values at each label in the training sample region and the test sample region obtained in step S2 are extracted according to the label position information given by the label, and a training sample set and a test sample set are obtained, specifically as follows:
[0068] 1. Use the "gdal.Open()" function to open the ROI data file in ENVI format;
[0069] 2. Use the "dataset.ReadAsArray()" function to read the multi-source remote sensing data into an array.
[0070] 3. Use the "np.argwhere()" function to find the index of all non-zero pixels.
[0071] 4. Use list comprehension to store the row index, column index and pixel value of these pixels into an array,
[0072] The above steps finally realize the reading and extraction of the feature values at each label in the training sample region and the test sample region.
[0073] In the embodiment of the application, the feature value is the pixel value of the corresponding pixel in the image, the remote sensing image has a spatial dimension, each channel represents a feature value, and different feature numbers are extracted according to different channel numbers, and the label information is extracted together to obtain a training sample set and a test sample set (.CSV format file).
[0074] Embodiment 2
[0075] The embodiment of the application provides a training sample set and a test sample set constructed according to the multi-source remote sensing data spatial sample set construction method of embodiment 1.
[0076] Embodiment 3
[0077] The embodiment of the application provides a multi-source remote sensing data information extraction method, which is realized by using the training sample set and the test sample set of embodiment 1, and includes the following steps:
[0078] S1, according to the requirement, a machine learning model is selected, different machine learning models are used for different requirements, a target machine learning model is obtained, and model parameters are set, the training sample set obtained in step S1 is input into the target machine learning model for model training, and a trained model is obtained;
[0079] The machine learning model in the embodiment of the application is any one of random forest (RF), support vector machine (SVM), decision tree (CART), multilayer perceptron (MLP), logistic (LOGISTIC), partial least squares (PLSR), xgboost, lightbm, catboost and adaboost. Some of the above models need to call different thread numbers according to the computer CPU performance, and some of the models need to select GPU acceleration according to the computer performance;
[0080] S2, verify the model accuracy of the trained model obtained in step S1 by using a confusion matrix on a test sample set, adjust the parameters of the trained model according to the verification result, and obtain an optimized model, specifically comprising the following steps:
[0081] S21, input the test sample set into the trained model obtained in step S1 to obtain a test sample prediction result, verify the prediction result of the test sample by using a confusion matrix to obtain an accuracy verification result, and the formula for accuracy verification is:
[0082]
[0083] Wherein, OA represents overall accuracy, TP represents true class, TN represents true negative class, FP represents false positive class, and FN represents false negative class; p o is the observed consistency, that is, the overall accuracy, p e is the expected consistency, that is, the accuracy of random classification.
[0084] OverallAccuracy (OA) overall accuracy and Kappa coefficient are indexes in the confusion matrix.
[0085] OverallAccuracy (OA) overall accuracy: it is the proportion of correct classification of the model in all classification samples, and it is an intuitive index for evaluating the overall performance of the model.
[0086] Kappa coefficient is an index for evaluating the performance of a classifier, considering the influence of accidental factors on the classification result, and the kappa coefficient provides a more objective evaluation index.
[0087] S22, according to the accuracy verification result obtained in step S21, adjust the parameters of the trained model to obtain an optimized model, specifically:
[0088] When the precision verification result OA and the Kappa coefficient obtained in step S21 are both greater than 0.8, it is determined that the trained model meets the requirements, and an optimized model is obtained; in the field of remote sensing image classification, an overall accuracy of 0.8 or above can be considered as a good model; and a Kappa coefficient of 0.8 or above can be considered as a good classification result.
[0089] When the precision verification result OA and the Kappa coefficient obtained in step S21 are both not greater than 0.8, it is determined that the trained model does not meet the requirements, the parameters of the model are adjusted, and then steps S1 and S2 are repeated until the precision verification result OA and the Kappa coefficient obtained in step S21 are both greater than 0.8, it is determined that the trained model meets the requirements, and an optimized model is obtained.
[0090] S3, input the target multi-source remote sensing data information into the optimized model obtained in step S5 for model prediction, and obtain a multi-source remote sensing data information extraction result.
[0091] Embodiment 4
[0092] As shown in Figures 4 to 8 , the embodiment of the present application provides a multi-source remote sensing data information extraction system, which is realized by the extraction method of embodiment 3, and the system of the embodiment of the present application is built based on PySide6, scikit-learn, and gdal, which is suitable for different remote sensing data and can achieve the purpose of quickly performing spatial sample random sampling and attribute feature value extraction of multi-source remote sensing data. Specifically, it includes:
[0093] A sample area creation module is used for pre-processing multi-source remote sensing images and creating a spatial total sample area; in the embodiment of the present application, the sample area creation module specifically includes:
[0094] A preprocessing module is used for pre-processing multi-source remote sensing images to obtain pre-processed images;
[0095] A sample creation module is used for selecting a target area on the pre-processed images and creating spatial sample areas of different categories on the target area;
[0096] A label assignment module is used for respectively assigning labels to spatial sample areas of different categories to obtain a spatial total sample area with label values.
[0097] The sample region division module (namely, a training and test sample division module) is used for randomly dividing a total spatial sample region into a spatial training sample region and a spatial test sample region; in the embodiment of the application, the total spatial sample region with a label value is selected, and a division ratio of 0.7 is input into the module, then the module is randomly divided into the spatial training sample region and the spatial test sample region according to a 7:3 ratio of the spatial training sample region to the spatial test sample region, saved to a specified path and named as **_ForTrain and **_ForText, and after completion, the system prompts.. / are splitted successfully. As shown in Figure 5 ;
[0098] The attribute feature extraction module is used for performing attribute feature extraction on the spatial training sample region and the spatial test sample region respectively to obtain a training sample set and a test sample set; in the embodiment of the application, the ForTrain and ForText sample region files (dat format files) obtained by the sample region division module are selected respectively, and corresponding target remote sensing images are selected, a saving path is set, target region image features are converted into vector features, and the vector features are extracted into corresponding sample sets (CSV format files) and saved to specified paths and named as **_ForTrain_pandas.csv and **_ForText_pandas.csv respectively, and after completion, the system prompts.. / are extracted successfully, as shown in Figure 6 ; for the obtained sample set (CSV format file), the file format is arranged from left to right as row number, column number, label value, feature value 1, feature value 2, …, feature value M, as shown in Figure 7 ;
[0099] The model training module is used for selecting a machine learning model according to requirements to obtain a target machine learning model, setting model parameters, inputting the training sample set into the target machine learning model for model training, and obtaining a trained model; in the embodiment of the application, taking a random forest model (RF) as an example, the obtained training sample set is selected, the number of decision trees is set to 200, the number of parallel threads is selected according to CPU performance, and the model is trained after the model position is saved, and after success, the system prompts ** / are saved successfully, as shown in Figure 8 ;
[0100] The model verification module (namely, an accuracy evaluation module) is used for verifying the accuracy of the trained model through the test sample set, adjusting the parameters of the trained model according to the verification result, and obtaining an optimized model; in the embodiment of the application, the random forest model and the data scaling ratio are selected, and the accuracy is evaluated, as shown in Figure 9The precision evaluation module in the embodiment of the application specifically comprises:
[0101] The evaluation module is configured to input the test sample set into the trained model to obtain a test sample prediction result, and verify the prediction result of the test sample by using a confusion matrix to obtain a precision verification result, wherein a formula for the precision verification is:
[0102]
[0103] wherein OA represents overall accuracy, TP represents a true class, TN represents a true negative class, FP represents a false positive class, and FN represents a false negative class; p o is observed agreement, i.e., overall accuracy, p e is expected agreement, i.e., accuracy of random classification.
[0104] The adjustment module is configured to adjust parameters of the trained model according to the precision verification result obtained by the evaluation module to obtain an optimized model, and the adjustment specifically comprises:
[0105] When the precision verification result OA and the Kappa coefficient obtained in step S21 are both greater than 0.8, it is determined that the trained model meets the requirements, and the optimized model is directly obtained.
[0106] When the precision verification result OA and the Kappa coefficient obtained in step S21 are both not greater than 0.8, it is determined that the trained model does not meet the requirements, the parameters of the model are adjusted, and then steps S1 and S2 are repeated until the precision verification result OA and the Kappa coefficient obtained in step S21 are both greater than 0.8, it is determined that the trained model meets the requirements, and the optimized model is obtained.
[0107] The data extraction module is configured to input target remote sensing image data into the optimized model for model prediction to obtain a remote sensing image data extraction result.
[0108] Although the embodiments of the application have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and purposes of the application, and the scope of the application is defined by the claims and their equivalents.
Claims
1. A method for constructing a spatial sample set of multi-source remote sensing data, characterized in that, Includes the following steps: Step 1: Preprocess the multi-source remote sensing images and create a total spatial sample region; Step 2: Randomly divide the total spatial sample region obtained in Step 1 into a spatial training sample region and a spatial test sample region; Step 3: Based on multi-source remote sensing data, extract attribute features from the spatial training sample area and spatial test sample area obtained in Step 2 to obtain the training sample set and the test sample set.
2. The method for constructing a spatial sample set of multi-source remote sensing data according to claim 1, characterized in that, Step 1 includes the following steps: Step 11: Preprocess the multi-source remote sensing images to obtain preprocessed images; Step 12: Select the target region on the preprocessed image obtained in step 11, and create spatial sample regions of different categories on the target region; Step 13: Assign labels to the different categories of spatial sample regions obtained in Step 12 to obtain the total spatial sample regions with label values.
3. The method for constructing a spatial sample set of multi-source remote sensing data according to claim 2, characterized in that, In step 12, the categories of the spatial sample region include regions of interest and regions of non-interest, and in step 13, the regions of interest are labeled according to their categories, while the labels of the regions of non-interest are assigned to 0.
4. The method for constructing a spatial sample set for multi-source remote sensing data according to claim 1, characterized in that, Step 2 includes the following steps: Based on the total spatial sample region obtained in step 1, the total spatial sample region obtained in step 1 is randomly divided into a spatial training sample region and a spatial test sample region according to the set division ratio.
5. The method for constructing a spatial sample set for multi-source remote sensing data according to claim 1, characterized in that, Step 3 includes the following steps: Based on multi-source remote sensing data, feature values at each label in the spatial training sample region and spatial test sample region obtained in step 2 are extracted to obtain the training sample set and the test sample set.
6. A training sample set and a test sample set constructed by the method for constructing a spatial sample set of multi-source remote sensing data according to any one of claims 1-5.
7. A method for extracting information from multi-source remote sensing data, implemented using the training sample set and test sample set as described in claim 6, characterized in that, Includes the following steps: S1. Select a machine learning model according to the requirements, obtain the target machine learning model, set the model parameters, input the training sample set into the target machine learning model for model training, and obtain the trained model. S2. Validate the accuracy of the trained model obtained in step S1 using a test sample set. Based on the validation results, adjust the parameters of the trained model to obtain an optimized model. S3. Input the target remote sensing image data into the optimized model obtained in step S2 to perform model prediction and obtain the remote sensing image information extraction results.
8. The method for extracting information from multi-source remote sensing data according to claim 7, characterized in that, In step S1, the machine learning model is any one of random forest, support vector machine, decision tree, multilayer perceptron, logistic, partial least squares, xgboost, lightbm, catboost and adaboost.
9. The method for extracting information from multi-source remote sensing data according to claim 7, characterized in that, Step S2 includes the following steps: S21. Input the test sample set into the trained model obtained in step S1 to obtain the prediction results of the test samples. Verify the accuracy of the prediction results of the test samples using a confusion matrix to obtain the accuracy verification results. The formula for accuracy verification is: Where OA represents the overall precision, TP represents the true positive class, TN represents the true negative class, FP represents the false positive class, and FN represents the false negative class; p o It refers to the observed consistency, i.e., the overall accuracy, p e It refers to the expected consistency, that is, the accuracy of random classification; S22. Based on the accuracy verification results obtained in step S21, adjust the parameters of the trained model to obtain the optimized model, specifically as follows: When the accuracy verification results OA and Kappa coefficient obtained in step S21 are both greater than 0.8, the trained model is determined to meet the requirements, and the optimized model is obtained. If the accuracy verification results OA and Kappa coefficient obtained in step S21 are both not greater than 0.8, the trained model is determined to be unacceptable. The model parameters are adjusted, and then steps S1 and S2 are repeated until the accuracy verification results OA and Kappa coefficient obtained in step S21 are both greater than 0.
8. Then the trained model is determined to be acceptable, and the optimized model is obtained.
10. A multi-source remote sensing data information extraction system, implemented by the method described in any one of claims 7-9, characterized in that, include: The sample region creation module is used to preprocess multi-source remote sensing images and create a total spatial sample region; The sample region partitioning module is used to randomly divide the total sample region into a spatial training sample region and a spatial test sample region. The attribute feature extraction module is used to extract attribute features from the spatial training sample region and the spatial test sample region respectively, so as to obtain the training sample set and the test sample set. The model training module is used to select a machine learning model according to the requirements, obtain the target machine learning model, set the model parameters, input the training sample set into the target machine learning model for model training, and obtain the trained model. The model validation module is used to validate the accuracy of the trained model using a test sample set. Based on the validation results, the parameters of the trained model are adjusted to obtain an optimized model. The data information extraction module is used to input the target remote sensing image data into the optimized model for model prediction and obtain the remote sensing image data information extraction results.