Remote sensing extraction method of water area multi-model fusion based on medium and high resolution remote sensing data

Abstract

The present invention discloses a multi-model fusion remote sensing extraction method for water areas of medium- and high-resolution remote sensing data, comprising the following extraction steps: S1, preparing high-resolution remote sensing images and medium- and high-resolution remote sensing images of the same area, extracting water bodies after selecting samples in the high-resolution remote sensing images, extracting water body data of the entire region in the medium- and high-resolution remote sensing images, checking and reviewing sample library data, performing calculations based on indicators, judging the accuracy of sample data, adjusting the sample data, optimizing the sample database, integrating multiple data algorithms, giving full play to the advantages of multi-source data, and increasing the resolution of water bodies in water areas, and providing a basis for extracting and classifying water body data in low-resolution remote sensing images by combining high-resolution and medium- and high-resolution sample data, thereby facilitating improving the accuracy of low-resolution remote sensing images.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology, and in particular to a water area multi-model fusion remote sensing extraction method for medium- and high-resolution remote sensing data. Background Art

[0002] When studying water areas, the most important thing is to extract water area data. Due to the terrain, area and difficulty of extraction, water area data is difficult to sample on a large scale like other data. Currently, remote sensing technology is widely used to obtain remote sensing map images and identify water areas in remote sensing images. When identifying water areas, the spectrum in the water area image is usually used. After identifying the water area data, the water area is segmented using image segmentation technology.

[0003] However, in the current water area extraction, the resolution of remote sensing images themselves is insufficient, which makes extraction difficult. There is no verification of multi-source data and the error rate is high. Summary of the Invention

[0004] The present invention provides a water area multi-model fusion remote sensing extraction method for medium- and high-resolution remote sensing data, which can effectively solve the problems raised in the above background technology in the current water area extraction, such as insufficient resolution of the remote sensing image itself, resulting in extraction difficulties, lack of multi-source data verification, and high error rate.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solution: a water area multi-model fusion remote sensing extraction method for medium- and high-resolution remote sensing data, comprising the following extraction steps:

[0006] S1. Prepare high-resolution remote sensing images and medium- and high-resolution remote sensing images of the same area;

[0007] S2. Select some samples from high-resolution remote sensing images and use AI to extract water bodies across the entire area;

[0008] S3. Perform global extraction of medium- and high-resolution remote sensing images using multiple remote sensing index methods, and continuously adjust the threshold to obtain water distribution results.

[0009] S4: Extract the same parts of the results of S2 and S3, manually check the results, add the qualified ones to the sample library of S2, and perform AI water area extraction of S2 again;

[0010] S5, randomly conducts manual review of the results, and those with correct results are added to the sample library of S2, and the AI ​​water area extraction of S2 is performed again;

[0011] S6, repeat iterative S5 until the manual inspection results are consistent with the real results of high-resolution remote sensing images;

[0012] S7, providing the results of S6 as samples to medium and high resolution data for training, and calculating the root mean square error, until the medium and high resolution data using machine learning extraction results reach the high resolution remote sensing image extraction results;

[0013] In S3 and S4, during the process of extracting water body data, the data classification is evaluated, and the evaluation indicators include: accuracy, precision, recall rate, F1Score value, misclassification error, omission error, kappa coefficient and ROC curve;

[0014] The calculation formula for accuracy is:

[0015]

[0016] In the formula, TP+TN is the number of samples that are correctly judged;

[0017] TP+FP+FN+TN is the total sample size;

[0018] The calculation formula for precision is:

[0019]

[0020] Where TP is the number of samples predicted correctly among the positive results predicted by the model;

[0021] TP+FP is the number of samples predicted as positive by the model;

[0022] The calculation formula for recall is:

[0023]

[0024] Where TP is the number of results whose true value is positive, and the model predicts the number correctly.

[0025] TP+FN is the number of samples predicted correctly by the model;

[0026] The calculation formula of F1Score is:

[0027]

[0028] In the formula, Precision is negatively correlated with Recall, and the F1 value range is [0, 1]. The closer it is to 1, the higher the accuracy of the model.

[0029] The calculation formula of commission error is:

[0030] Misclassification error + accuracy = 1(5);

[0031] The calculation formula for omission error (OmissionError) is:

[0032] Missing error + recall rate = 1(6);

[0033] The calculation formula of Kappa Coefficient is as follows:

[0034]

[0035] Where Pe is the ratio of (TP+FN)*(TP+FP)+(FP+TN)*(FN+TN) to the total number of elements;

[0036] Among them, (TP+FN) represents the number of samples judged as positive in the original data set;

[0037] (FP+TN) represents the number of samples judged as negative in the original data set;

[0038] (TP+FP) represents the number of samples judged as positive after classification;

[0039] (FN+TN) represents the number of samples judged as negative after classification;

[0040] The calculated result of the kappa coefficient is between [-1, 1], and the closer it is to 1, the higher the consistency;

[0041] The Y-axis of the ROC curve (Receiver Operating Characteristic curve) represents the true positive rate (TPR), TPR is the recall rate, and the X-axis represents the false positive rate (FPR), FPR is the probability that a negative example is reported as a positive example, that is, the omission error;

[0042] The ROC curve of the model is quantified by calculating the area under the curve (AUC). The value range of AUC is [0, 1]. The AUC of a random classifier is 0.5. The larger the AUC value of the model, the better the trade-off between accuracy and recall rate is achieved, indicating that the model classification performance is better.

[0043] According to the above technical solution, the specific steps of extracting the water area in S2 are as follows:

[0044] Sample data preparation: obtain remote sensing image data for annotation, convert it into a single-channel mask file of the same size as the image data, and slice the image and mask file;

[0045] Model training: Using the U-Net network architecture, based on a fully convolutional network (FCN), the encoder consists of convolution and downsampling modules, and the decoder consists of upsampling deconvolution and feature concatenation. The data is fused and a feature map is generated after two convolutions. Then, based on the specific classification results, n 1x1 convolutions are used for classification to obtain heatmaps for each category. Finally, the softmax function is used to determine the most likely category, thus completing the final segmentation operation.

[0046] Result prediction: After training, the model parameters are saved as a file, and the same cropping and reading operations are performed on the predicted image. After the prediction is completed, the prediction results are spliced ​​to obtain the prediction results of the entire domain;

[0047] Post-processing: Further process the segmentation results to enhance the display effect and remove existing errors.

[0048] According to the above technical solution, in S2, high-resolution data water body identification is to extract water bodies from sample data and then take the intersection to obtain raster data of medium- and high-resolution index-extracted water areas, where the remote sensing index formula for extracting water areas is:

[0049] NDWI=(Green-NIR) / (Green+NIR)(8);

[0050] EWI=(Green-NIR-MIR) / (Green+NIR+MIR)(9);

[0051] MDNWI=(Green-MIR) / (Green+MIR)(10).

[0052] According to the above technical solution, in S4 and S5, the verified accurate data is added to the sample library of S2 through manual interpretation, and the inaccurate data is verified, and a part of the results are extracted by machine recognition as samples, which are mixed with the results of manual interpretation and trained again;

[0053] The algorithm results are verified for authenticity through ground truth collected manually and by instruments in the field, and the results are finally corrected;

[0054] After the sample library is determined, the data is classified, and then the classification results are evaluated by comparing the degree of conformity between the test samples and the data. Before the evaluation, the symbols are defined:

[0055] The set of samples that are positive in the original data set and still positive after classification is recorded as TP;

[0056] The set of samples that belong to the positive class in the original data set but belong to the negative class after classification is recorded as FN;

[0057] The set of samples that belong to the negative class in the original data set but are classified as the positive class after classification is recorded as FP;

[0058] The set of samples that belong to the negative class in the original data set and are still negative class after classification is recorded as TN;

[0059] The more TPs and TNs there are and the fewer FPs and FNs there are, the higher the classification accuracy of the model, which is also called the confusion matrix.

[0060] According to the above technical solution, in S7, a regression equation is established to predict the target value. Commonly used indicators include: mean square error, root mean square error, mean absolute error and determination coefficient;

[0061] The mean square error (MSE) calculation formula is as follows:

[0062]

[0063] In the formula, predicti-truthi is the deviation between the predicted value and the true value;

[0064] n is the total amount of data;

[0065] The smaller the MSE value, the better the fitting effect;

[0066] The root mean square error (RMSE) is calculated as follows:

[0067]

[0068] In the formula, predicti-truthi is the deviation between the predicted value and the true value;

[0069] n is the total amount of data;

[0070] The mean absolute error (MAE) is calculated as follows:

[0071]

[0072] Where predicti-truthi is the deviation between the predicted value and the true value;

[0073] n is the total amount of data;

[0074] The coefficient of determination (R2) is calculated as follows:

[0075]

[0076] Its value range is [0, 1]. The closer it is to 1, the more the independent variable can explain the variance change of the dependent variable. The smaller the value, the worse the effect.

[0077] According to the above technical solution, in S7, the U-Net model is used to train the medium and high resolution remote sensing data multiple times to extract water area sample segmentation training.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] By selecting samples in high-resolution remote sensing images and then extracting water bodies, extracting water body data for the entire region in medium- and high-resolution remote sensing images, and checking and reviewing the sample library data, calculating according to indicators, judging the accuracy of sample data, adjusting the sample data, and optimizing the sample database, multiple data algorithms are integrated to give full play to the advantages of multi-source data and increase the resolution of water bodies in water areas. By combining high-resolution and medium- and high-resolution sample data, a basis is provided for the extraction and classification of water body data in low-resolution remote sensing images, thereby facilitating the improvement of the accuracy of low-resolution remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0081] Figure 1 is a flow chart of the extraction steps of the present invention;

[0082] Figure 2 It is a schematic diagram of remote sensing image of the present invention;

[0083] Figure 3 It is a schematic diagram of a mask file corresponding to the remote sensing image of the present invention;

[0084] Figure 4 It is a schematic diagram of the U-Net network structure of the present invention;

[0085] Figure 5 Schematic diagram comparing the prediction results of the present invention with the labels;

[0086] Figure 6 2. It is a schematic diagram of the result of eliminating holes in the prediction results according to the present invention;

[0087] Figure 7 This is a diagram of the accuracy training process of the U-Net model of the present invention;

[0088] Figure 8 This is a diagram of the error training process of the U-Net model of the present invention;

[0089] Figure 9 It is a schematic diagram of the local prediction result of the present invention;

[0090] Figure 10 It is a schematic diagram of the overall prediction results of the present invention. DETAILED DESCRIPTION

[0091] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0092] Example: Figure 1 As shown, the present invention provides a technical solution, a water area multi-model fusion remote sensing extraction method for medium and high resolution remote sensing data, comprising the following extraction steps:

[0093] S1. Prepare high-resolution remote sensing images and medium- and high-resolution remote sensing images of the same area;

[0094] S2. Select some samples from high-resolution remote sensing images and use AI to extract water bodies across the entire area;

[0095] S3. Perform global extraction of medium- and high-resolution remote sensing images using multiple remote sensing index methods, and continuously adjust the threshold to obtain water distribution results.

[0096] S4: Extract the same parts of the results of S2 and S3, manually check the results, add the qualified ones to the sample library of S2, and perform AI water area extraction of S2 again;

[0097] S5, randomly conducts manual review of the results, and those with correct results are added to the sample library of S2, and the AI ​​water area extraction of S2 is performed again;

[0098] S6, repeat iterative S5 until the manual inspection results are consistent with the real results of high-resolution remote sensing images;

[0099] S7, providing the results of S6 as samples to medium and high resolution data for training, and calculating the root mean square error, until the medium and high resolution data using machine learning extraction results reach the high resolution remote sensing image extraction results;

[0100] In S3 and S4, during the process of extracting water body data, the data classification is evaluated. The evaluation indicators include: accuracy, precision, recall, F1Score value, misclassification error, omission error, kappa coefficient and ROC curve;

[0101] The calculation formula for accuracy is:

[0102]

[0103] In the formula, TP+TN is the number of samples that are correctly judged;

[0104] TP+FP+FN+TN is the total sample size;

[0105] The calculation formula for precision is:

[0106]

[0107] Where TP is the number of samples predicted correctly among the positive results predicted by the model;

[0108] TP+FP is the number of samples predicted as positive by the model;

[0109] The calculation formula for recall is:

[0110]

[0111] Where TP is the number of results whose true value is positive, and the model predicts the number correctly.

[0112] TP+FN is the number of samples predicted correctly by the model;

[0113] The calculation formula of F1Score is:

[0114]

[0115] In the formula, Precision is negatively correlated with Recall, and the F1 value range is [0, 1]. The closer it is to 1, the higher the accuracy of the model.

[0116] The calculation formula of commission error is:

[0117] Misclassification error + accuracy = 1(5);

[0118] The calculation formula for omission error (OmissionError) is:

[0119] Missing error + recall rate = 1(6);

[0120] The calculation formula of Kappa Coefficient is as follows:

[0121]

[0122] Where Pe is the ratio of (TP+FN)*(TP+FP)+(FP+TN)*(FN+TN) to the total number of elements;

[0123] Among them, (TP+FN) represents the number of samples judged as positive in the original data set;

[0124] (FP+TN) represents the number of samples judged as negative in the original data set;

[0125] (TP+FP) represents the number of samples judged as positive after classification;

[0126] (FN+TN) represents the number of samples judged as negative after classification;

[0127] The calculated result of the kappa coefficient is between [-1, 1], and the closer it is to 1, the higher the consistency;

[0128] The Y-axis of the ROC curve (Receiver Operating Characteristic curve) represents the true positive rate (TPR), TPR is the recall rate, and the X-axis represents the false positive rate (FPR), FPR is the probability that a negative example is reported as a positive example, that is, the omission error;

[0129] The ROC curve of the model is quantified by calculating the area under the curve (AUC). The value range of AUC is [0, 1]. The AUC of a random classifier is 0.5. The larger the AUC value of the model, the better the trade-off between accuracy and recall rate is achieved, indicating that the model classification performance is better.

[0130] According to the above technical solution, the specific steps for extracting the water area in S2 are as follows:

[0131] like Figure 2-3 As shown, sample data preparation: remote sensing image data is obtained for annotation, converted into a single-channel mask file of the same size as the image data, and the image and mask file are sliced;

[0132] like Figure 4 As shown in the figure, model training: using the U-Net network architecture, based on the fully convolutional network (FCN), the encoder is composed of convolution and downsampling modules, and the decoder is composed of upsampling deconvolution and feature splicing. The data is fused and a feature map is generated after two convolutions. Then, according to the specific classification results, n 1x1 convolution classification operations are used to obtain the heatmap of each category. Finally, the softmax function is used to determine the most likely category, thereby completing the final segmentation operation.

[0133] like Figure 5 As shown in the figure, the result prediction: after the training, the model parameters are saved as a file, and the same cropping and reading operations are performed on the predicted image. After the prediction is completed, the prediction results are spliced ​​to obtain the prediction results of the entire domain;

[0134] like Figure 6 As shown, post-processing: further process the segmentation results to enhance the display effect of the results and remove existing errors.

[0135] According to the above technical solution, in S2, high-resolution data water body identification is to extract water bodies from sample data and then take the intersection to obtain raster data of medium- and high-resolution index-extracted water areas. The remote sensing index formula for extracting water areas is:

[0136] NDWI=(Green-NIR) / (Green+NIR)(8);

[0137] EWI=(Green-NIR-MIR) / (Green+NIR+MIR)(9);

[0138] MDNWI=(Gteen-MIR) / (Green+MIR)(10).

[0139] According to the above technical solution, in S4 and S5, the verified accurate data is added to the sample library of S2 through manual interpretation. For the inaccurate data, a part of the results is extracted by machine recognition as samples and mixed with the results of manual interpretation for retraining;

[0140] The algorithm results are verified for authenticity through ground truth collected manually and by instruments in the field, and the results are finally corrected;

[0141] After the sample library is determined, the data is classified, and then the classification results are evaluated by comparing the degree of conformity between the test samples and the data. Before the evaluation, the symbols are defined:

[0142] The set of samples that are positive in the original data set and still positive after classification is recorded as TP;

[0143] The set of samples that belong to the positive class in the original data set but belong to the negative class after classification is recorded as FN;

[0144] The set of samples that belong to the negative class in the original data set but are classified as the positive class after classification is recorded as FP;

[0145] The set of samples that belong to the negative class in the original data set and are still negative class after classification is recorded as TN;

[0146] The more TPs and TNs there are and the fewer FPs and FNs there are, the higher the classification accuracy of the model will be. This is also called the confusion matrix, as shown in the following table:

[0147]

[0148] According to the above technical solution, in S7, a regression equation is established to predict the target value. Commonly used indicators include: mean square error, root mean square error, mean absolute error and determination coefficient;

[0149] The mean square error (MSE) calculation formula is as follows:

[0150]

[0151] In the formula, predicti-truthi is the deviation between the predicted value and the true value;

[0152] n is the total amount of data;

[0153] The smaller the MSE value, the better the fitting effect;

[0154] The root mean square error (RMSE) is calculated as follows:

[0155]

[0156] In the formula, predicti-truthi is the deviation between the predicted value and the true value;

[0157] n is the total amount of data;

[0158] The mean absolute error (MAE) is calculated as follows:

[0159]

[0160] Where predicti-truthi is the deviation between the predicted value and the true value;

[0161] n is the total amount of data;

[0162] The coefficient of determination (R2) is calculated as follows:

[0163]

[0164] Its value range is [0, 1]. The closer it is to 1, the more the independent variable can explain the variance change of the dependent variable. The smaller the value, the worse the effect.

[0165] like Figure 7-10 As shown, according to the above technical solution, in S7, the U-Net model is used to train the medium and high resolution remote sensing data multiple times to extract water sample segmentation training

[0166] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A water area multi-model fusion remote sensing extraction method for medium- and high-resolution remote sensing data, characterized by: The extraction steps include: S1. Prepare high-resolution remote sensing images and medium- and high-resolution remote sensing images of the same area; S2. Select some samples from high-resolution remote sensing images and use AI to extract water bodies across the entire area; S3. Perform global extraction of medium- and high-resolution remote sensing images using multiple remote sensing index methods, and continuously adjust the threshold to obtain water distribution results. S4: Extract the same parts of the results of S2 and S3, manually check the results, add the qualified ones to the sample library of S2, and perform AI water area extraction of S2 again; S5, randomly conducts manual review of the results, and those with correct results are added to the sample library of S2, and the AI ​​water area extraction of S2 is performed again; S6, repeat iterative S5 until the manual inspection results are consistent with the real results of high-resolution remote sensing images; S7, providing the results of S6 as samples to medium and high resolution data for training, and calculating the root mean square error, until the medium and high resolution data using machine learning extraction results reach the high resolution remote sensing image extraction results; In S3 and S4, during the process of extracting water body data, the data classification is evaluated, and the evaluation indicators include: accuracy, precision, recall rate, F1Score value, misclassification error, omission error, kappa coefficient and ROC curve; The calculation formula for accuracy is: In the formula, TP+TN is the number of samples that are correctly judged; TP+FP+FN+TN is the total sample size; The calculation formula for precision is: Where TP is the number of samples predicted correctly among the positive results predicted by the model; TP+FP is the number of samples predicted as positive by the model; The calculation formula for recall is: Where TP is the number of results whose true value is positive, and the model predicts the number correctly. TP+FN is the number of samples predicted correctly by the model; The calculation formula of F1Score is: In the formula, Precision is negatively correlated with Recall, and the F1 value range is [0, 1]. The closer it is to 1, the higher the accuracy of the model. The calculation formula for misclassification error is: Misclassification error + accuracy = 1(5); The calculation formula for omission error is: Missing error + recall rate = 1(6); The calculation formula of the kappa coefficient is as follows: Where Pe is the ratio of (TP+FN)*(TP+FP)+(FP+TN)*(FN+TN) to the total number of elements; Among them, (TP+FN) represents the number of samples judged as positive in the original data set; (FP+TN) represents the number of samples judged as negative in the original data set; (TP+FP) represents the number of samples judged as positive after classification; (FN+TN) represents the number of samples judged as negative after classification; The calculated result of the kappa coefficient is between [-1, 1], and the closer it is to 1, the higher the consistency; The Y-axis of the ROC curve represents the true positive rate (TPR), TPR is the recall rate, and the X-axis represents the false positive rate (FPR), FPR is the probability that a negative example is reported as a positive example, that is, the omission error; The ROC curve of the model is quantified by calculating the area under the curve (AUC). The value range of AUC is [0, 1]. The AUC of a random classifier is 0.

5. The larger the AUC value of the model, the better.

2. The method for extracting water area multi-model fusion remote sensing data from medium and high resolution remote sensing data according to claim 1, characterized in that: The specific steps of extracting the water area in S2 are as follows: Sample data preparation: obtain remote sensing image data for annotation, convert it into a single-channel mask file of the same size as the image data, and slice the image and mask file; Model training: Using the U-Net network architecture, based on a fully convolutional network (FCN), the encoder consists of convolution and downsampling modules, and the decoder consists of upsampling deconvolution and feature concatenation. The data is fused and a feature map is generated after two convolutions. Then, based on the specific classification results, n 1x1 convolutions are used for classification to obtain heatmaps for each category. Finally, the softmax function is used to determine the most likely category, thus completing the final segmentation operation. Result prediction: After training, the model parameters are saved as a file, and the same cropping and reading operations are performed on the predicted image. After the prediction is completed, the prediction results are spliced ​​to obtain the prediction results of the entire domain; Post-processing: Further process the segmentation results to enhance the display effect and remove existing errors.

3. The method for extracting water area multi-model fusion remote sensing data from medium and high resolution remote sensing data according to claim 1, characterized in that: In S2, high-resolution data water body identification is to extract water bodies from sample data and then take the intersection to obtain raster data of water areas with medium and high-resolution index extraction. The remote sensing index formula for extracting water areas is: NDWI=(Green-NIR) / (Green+NIR) (8); EWI=(Green-NIR-MIR) / (Green+NIR+MIR) (9); MDNWI=(Green-MIR) / (Green+MIR) (10).

4. The method for extracting water area multi-model fusion remote sensing data from medium and high resolution remote sensing data according to claim 1, characterized in that: In S4 and S5, the verified accurate data is added to the sample library of S2 through manual interpretation, and the inaccurate data is verified, and a part of the results are extracted by machine recognition as samples, which are mixed with the manual interpretation results for retraining; The algorithm results are verified for authenticity through ground truth collected manually and by instruments in the field, and the results are finally corrected; After the sample library is determined, the data is classified, and then the classification results are evaluated by comparing the degree of conformity between the test samples and the data. Before the evaluation, the symbols are defined: The set of samples that are positive in the original data set and still positive after classification is recorded as TP; The set of samples that belong to the positive class in the original data set but belong to the negative class after classification is recorded as FN; The set of samples that belong to the negative class in the original data set but are classified as the positive class after classification is recorded as FP; The set of samples that belong to the negative class in the original data set and are still negative class after classification is recorded as TN; The more TPs and TNs there are and the fewer FPs and FNs there are, the higher the classification accuracy of the model, which is also called the confusion matrix.

5. The method for extracting water area multi-model fusion remote sensing data from medium and high resolution remote sensing data according to claim 1, characterized in that: In said S7, a regression equation is established to predict the target value, and commonly used indicators include: mean square error, root mean square error, mean absolute error and determination coefficient; The mean square error (MSE) is calculated as follows: In the formula, predicti-truthi is the deviation between the predicted value and the true value; n is the total amount of data; The smaller the MSE value, the better the fitting effect; The root mean square error (RMSE) is calculated as follows: In the formula, predicti-truthi is the deviation between the predicted value and the true value; n is the total amount of data; The mean absolute error (MAE) is calculated as follows: Where predicti-truthi is the deviation between the predicted value and the true value; n is the total amount of data; The coefficient of determination (R2) is calculated as follows: Its value range is [0, 1]. The closer it is to 1, the more the independent variable can explain the variance change of the dependent variable. The smaller the value, the worse the effect.

6. The method for extracting water area multi-model fusion remote sensing data from medium and high resolution remote sensing data according to claim 1, characterized in that: In S7, the U-Net model is used to train the medium and high resolution remote sensing data multiple times to extract water sample segmentation training.

Images