Artificial forest multi-source remote sensing extraction method
Through the preprocessing of multi-source remote sensing data and the decision-making fusion of classifiers, combined with NDVI processing and harmonic regression fitting, the problem of low classification accuracy and stability of artificial forests in the existing technology is solved, and more efficient classification results are achieved .
Patent Information
- Application Number
- CN202411201582.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-29
AI Technical Summary
In the classification of plantation and natural forests, the accuracy and stability of classification results are not high due to the influence of data noise and the inability to fully utilize the multi-dimensional characteristics of multi-source remote sensing data.
The multi-source remote sensing data extraction method is adopted, including obtaining multi-source remote sensing data for preprocessing, training multiple classifiers and performing sample enhancement, generating final classification results through decision fusion, and extracting time series features using NDVI processing and harmonic regression fitting. Integrates random forests, decision trees and support vector machine classifiers.
It improves the accuracy and stability of plantation classification, enhances the ability to identify a few types of samples, overcomes the deviation problem of a single classifier, and improves the comprehensive accuracy and robustness of the final classification results.
Smart Images

Figure CN119091301A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of remote sensing data detection, and in particular relates to a multi-source remote sensing extraction method for artificial forests. Background Art
[0002] With the rapid development of remote sensing technology and geographic information systems, the use of remote sensing data for large-scale forest monitoring and classification has become an important means of forestry management. Remote sensing technology provides reliable data support for studying the spatial distribution, dynamic changes and relationship between forest resources and environmental factors by acquiring large-scale and high-frequency surface information. In particular, with the popularization of high-resolution and multi-spectral satellite images, remote sensing technology has been widely used in forest classification, forest health monitoring, carbon storage assessment and other fields.
[0003] Time series remote sensing data, that is, observing the changes of surface targets over time through continuous multi-temporal images, has become an important means in vegetation monitoring research. By analyzing the time series changes of the Normalized Difference Vegetation Index (NDVI), the growth cycle of vegetation, seasonal changes, and the impact of abnormal events on vegetation can be effectively captured. Especially in forest classification, time series analysis can identify and distinguish the growth patterns and seasonal characteristics of different types of forests, thereby improving the accuracy of classification. Modern remote sensing technology does not only rely on a single type of image data, but has gradually developed methods for the fusion of multi-source remote sensing data, such as multi-source data including optical images, radar data, and digital elevation models. These data types have their own advantages in spatial resolution, spectral resolution, and temporal resolution. By fusing multi-source data, more abundant surface information can be obtained. In forest classification, multi-source data fusion technology can make full use of spectral, spatial structure, and terrain information to enhance the ability to identify forest types from different dimensions. At the same time, with the improvement of computing power and the advancement of machine learning algorithms, remote sensing image classification technology based on machine learning has become a hot topic of research. Random Forest, Support Vector Machine (SVM) and deep learning algorithms have shown strong capabilities in processing high-dimensional remote sensing data. Especially when dealing with complex terrain and the coexistence of multiple objects, machine learning algorithms can effectively utilize the characteristics of training samples to build a robust classification model.
[0004] In existing studies, the classification algorithm for distinguishing similar types of land objects combines long time series data with machine learning methods to improve classification accuracy. For example, time series analysis based on the Harmonic Analysis of Time Series (HANTS) algorithm can remove the influence of clouds through harmonic analysis, reconstruct remote sensing time series data, and use time series trend characteristics to distinguish different types of forests. The application of these studies shows that the application of long time series data can significantly improve the accuracy of artificial forest identification and reduce uncertainty in classification. However, due to the significant difference between the number of artificial forest and natural forest samples, the complex forest classification processing task is greatly affected by data noise, and the multidimensional features in multi-source remote sensing data cannot be fully utilized during the processing process, resulting in low comprehensive accuracy and stability of the final classification results. Therefore, a multi-source remote sensing extraction method for artificial forests is needed to improve the accuracy and stability of classification results. Summary of the invention
[0005] In order to solve the above problems existing in the prior art, the present invention provides a method for multi-source remote sensing extraction of artificial forests. The technical problem to be solved by the present invention is achieved through the following technical solutions:
[0006] The present invention provides a method for extracting artificial forest multi-source remote sensing data, comprising:
[0007] Acquire multi-source remote sensing data;
[0008] Preprocessing the multi-source remote sensing data to obtain preprocessed multi-source remote sensing data;
[0009] Obtain training samples and verification samples respectively, and filter the training samples according to the verification samples to obtain a first sample set;
[0010] Training a plurality of classifiers according to the first sample set to obtain a plurality of first classifiers, and classifying the first sample set by the plurality of first classifiers to obtain a plurality of first classification results;
[0011] Performing sample enhancement on a minority of the first classification results to obtain a second sample set;
[0012] Several first classifiers are trained respectively according to the second sample set to obtain several second classifiers accordingly; the pre-processed multi-source remote sensing data are classified respectively by the several second classifiers to obtain several second classification results accordingly, and the several second classification results are fused by decision making to obtain the final classification result.
[0013] In one embodiment of the present invention, acquiring multi-source remote sensing data includes: acquiring Sentinel-2 time series images, Sentinel-1 time series images, and Landsat time series images respectively.
[0014] In one embodiment of the present invention, preprocessing the multi-source remote sensing data to obtain preprocessed multi-source remote sensing data includes:
[0015] De-clouding was performed on Sentinel-2 time series images, Sentinel-1 time series images, and Landsat time series images to obtain multispectral band information.
[0016] The declouded Landsat time series images were processed by NDVI to obtain the Normalized Difference Vegetation Index;
[0017] Perform harmonic regression fitting on the normalized vegetation index and extract statistical characteristics of the fitted normalized vegetation index; wherein the statistical characteristics include: mean, variance, standard deviation, coefficient of variation and corrected variance;
[0018] Multi-source remote sensing data also includes: terrain feature information and radar image data information; multi-source feature integration is performed on multi-spectral band information, statistical features, terrain feature information and radar image data information to obtain an integrated result;
[0019] The integrated results are extracted to obtain texture features.
[0020] In one embodiment of the present invention, the formula for NDVI processing is:
[0021]
[0022] Among them, NDVI is the normalized vegetation index; NIR is the reflectance value of the near-infrared band; Red is the reflectance value of the red light band.
[0023] In one embodiment of the present invention, the model of harmonic regression fitting is:
[0024]
[0025] in, is the set of normalized vegetation indices after fitting reconstruction; n is the number of harmonics; t j is the time step, j = 1, 2, ..., N, N is the number of observations in the time series; p i and q i are the coefficients of the trigonometric functions in the Fourier series, f i is the corresponding frequency; x0 is a constant term.
[0026] In one embodiment of the present invention, respectively obtaining training samples and verification samples, and screening the training samples according to the verification samples to obtain a first sample set includes:
[0027] Obtain training samples and validation samples respectively, where the validation samples are forest layer data;
[0028] Compare the training samples with the forest layer data, and select the training samples in the forest layer data as the reserved samples;
[0029] Labels are reprojected on the retained samples to obtain the first sample set.
[0030] In one embodiment of the present invention, several classifiers are trained according to the first sample set to obtain several first classifiers, and the first sample set is classified by the several first classifiers to obtain first classification results, including:
[0031] Select minority class samples and majority class samples from the first sample set as training sample sets respectively;
[0032] Several classifiers are trained respectively according to the training sample sets; wherein the several classifiers include: random forest classifier, decision tree classifier and support vector machine classifier;
[0033] Adjust the proportion of minority class samples in the training sample set, and retrain several classifiers according to the adjusted training sample set until the evaluation index is met, thereby obtaining several first classifiers; wherein the evaluation index is used to evaluate the classification accuracy of the several trained classifiers;
[0034] The first sample set is classified according to a plurality of first classifiers respectively, and corresponding first classification results are obtained.
[0035] In one embodiment of the present invention, the evaluation indicators include: confusion matrix results, overall precision and recall;
[0036] Confusion matrix results, including true positives TP, false positives FP, false negatives FN, and true negatives TN, are used to determine the classification accuracy of several classifiers;
[0037] The formula for overall accuracy is:
[0038]
[0039] Among them, Overall Accuracy is the overall accuracy;
[0040] The formula for recall is:
[0041]
[0042] Among them, Recall is the recall rate.
[0043] In one embodiment of the present invention, performing sample enhancement on a minority of the first classification results to obtain a second sample set includes:
[0044] A few of the first classification results are overlapped with the forest layer data to obtain the overlapped area, which is:
[0045] A overlap =A s ∩A f ;
[0046] Among them, A overlap A is the overlap area between the minority classification results and the forest layer data; s A is the area of minority classification results; f The area for the forest layer data;
[0047] A buffer is created with the minority classification results located in the overlapping area as the center, new sample points are generated in the buffer to achieve sample enhancement, and a second sample set is obtained based on the enhanced samples;
[0048] The formula for the buffer is:
[0049] B i ={(x,y)|(x i ) 2 +(yy i ) 2 ≤r 2};
[0050] Where (x, y) is any point in the buffer zone; p i is the minority classification result, and its coordinates are (x i ,y i );B i is the minority classification result p i The buffer zone with the center being the circle and the radius being r;
[0051] The formula for the new sample point is:
[0052]
[0053] Among them, p new is the new sample point generated in the buffer; (x j ,y j ) is the coordinate of the new sample point generated in the buffer; n new The number of new sample points generated in the buffer.
[0054] In one embodiment of the present invention, several first classifiers are trained respectively according to the second sample set, and several second classifiers are obtained accordingly, including:
[0055] Training a plurality of first classifiers respectively according to the second sample set, and obtaining a plurality of second classifiers accordingly;
[0056] Classifying the preprocessed multi-source remote sensing data according to a plurality of second classifiers, and obtaining a plurality of second classification results accordingly;
[0057] The decision fusion of several second classification results is performed to obtain the final classification result, which includes:
[0058]
[0059] in, is the final classification result; z is the pixel point; mode is the function for calculating the category with the most occurrences among the three classification results; and The following are the classification results of several second classifiers.
[0060] Compared with the prior art, the present invention has the following beneficial effects:
[0061] The artificial forest multi-source remote sensing extraction method of the present invention is based on multi-source remote sensing data, performs sample enhancement according to the first classification result of the trained first classifier, and generates new sample points in the buffer zone of the minority class sample points, which not only effectively increases the number of minority class samples, but also ensures the reasonable distribution of the new sample points, improves the classifier's recognition ability for minority class samples, and solves the problem of class imbalance.
[0062] The present invention utilizes the time series characteristics of NDVI processing and extracts the trend and seasonal change characteristics in the time series through harmonic regression fitting, which not only enhances the classifier's ability to capture vegetation changes, but also improves the classifier's sensitivity to long-term changes and periodic changes, providing support for obtaining more accurate classification results.
[0063] The present invention trains three different classifiers, random forest, decision tree and support vector machine, and performs decision fusion to generate the final classification result. The fusion strategy can effectively integrate the advantages of different classifiers, overcome the bias problem that is prone to occur in a single classifier, and improve the comprehensive accuracy, stability and robustness of the final classification result.
[0064] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following specifically cites a preferred embodiment and describes it in detail with the accompanying drawings as follows. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 It is a method flow chart of a method for multi-source remote sensing extraction of artificial forests provided by an embodiment of the present invention;
[0066] Figure 2 It is a workflow diagram of a multi-source remote sensing extraction method for artificial forests provided by an embodiment of the present invention;
[0067] Figure 3 is a workflow diagram of a sample enhancement method provided by an embodiment of the present invention;
[0068] Figure 4 It is a workflow diagram of the decision fusion method provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0069] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, a multi-source remote sensing extraction method for artificial forests proposed by the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0070] The above and other technical contents, features and effects of the present invention are clearly presented in the following detailed description of the specific implementation modes in conjunction with the accompanying drawings. Through the description of the specific implementation modes, the technical means and effects adopted by the present invention to achieve the predetermined purpose can be more deeply and specifically understood. However, the attached drawings are only for reference and explanation purposes and are not used to limit the technical solutions of the present invention.
[0071] Embodiment 1
[0072] like Figure 1 and Figure 2 As shown, this embodiment provides a method for extracting artificial forests by multi-source remote sensing, including:
[0073] S1: Acquire multi-source remote sensing data;
[0074] In an optional implementation, S1 includes: acquiring Sentinel-2 time series images, Sentinel-1 time series images, Landsat time series images, multispectral remote sensing bands, terrain features and radar image data respectively.
[0075] S2: preprocessing the multi-source remote sensing data to obtain preprocessed multi-source remote sensing data;
[0076] In an optional embodiment, S2 includes:
[0077] S2.1: Decloud the Sentinel-2 time series images, Sentinel-1 time series images and Landsat time series images respectively to obtain multispectral band information;
[0078] Exemplarily, the de-clouding process of Sentinel-2 time series images includes: selecting Sentinel-2 images of 12 bands including B1-B12 from the COPERNICUS / S2_SR dataset of the GEE (Google EarthEngine) platform; using the COPERNICUS / S2_CLOUD_PROBABILITY dataset, according to the set cloud probability threshold (65%), masking the cloud-covered parts of the selected 12 bands of Sentinel-2 images, and retaining pixels with cloud coverage below the threshold; for edge noise problems that may exist in some 10-meter resolution bands, applying 20-meter and 60-meter band masks to process edge data to ensure image quality; setting the time range of the image (January 1, 2018 to December 31, 2020) and the geometric range of the study area, and screening the image. The band parameters of the Sentinel-2 time series images are shown in Table 1.
[0079] Table 1 Band parameters of Sentinel-2 time series images
[0080]
[0081] It is worth noting that the multi-source remote sensing extraction method for plantations in this embodiment realizes large-scale remote sensing data processing and analysis based on the GEE platform. The powerful computing power and data storage capacity of the GEE platform enable the multi-source remote sensing extraction method for plantations in this embodiment to perform efficient classification and analysis based on large-scale and long-time series data. Compared with traditional local processing methods, it significantly improves data processing efficiency and reduces hardware costs and time consumption.
[0082] Exemplarily, declouding of Sentinel-1 time series images includes: selecting Sentinel-1 time series images containing VV and VH polarization information from the COPERNICUS / S1_GRD dataset; masking the noise portion of the image using a set threshold to exclude abnormal reflection values; and performing median synthesis processing on the images within the time range to reduce the interference of outliers in the time series and ensure the uniformity of the image.
[0083] Exemplarily, declouding Landsat time series images includes: selecting Landsat4, 5, 7 and 8 images from the Landsat series image collection, filtering them according to the area of interest and time range; applying corresponding scaling factors to the optical band and thermal infrared band to ensure the physical consistency of the Landsat image; linearly scaling and offset correction of the band values according to the specific requirements of each Landsat series image to obtain the actual surface reflectance and temperature values; using the quality assessment band (QA_PIXEL) in the image to generate a cloud mask, and excluding some pixels covered by clouds, cloud shadows and snow, and retaining valid data without cloud coverage. The above processing ensures that the pixels in the image are free of interference, thereby improving the analysis quality of the image.
[0084] S2.2: Perform NDVI processing on the declouded Landsat time series images to obtain the Normalized Difference Vegetation Index;
[0085] In an optional embodiment, the declouded Landsat time series images are subjected to NDVI processing, and the normalized vegetation index is obtained by selecting the near infrared band and the red light band. Subsequently, a time variable (in years, starting from 1970) is added to the image for subsequent time series processing.
[0086] Specifically, the formula for NDVI processing is:
[0087]
[0088] Among them, NDVI is the normalized vegetation index; NIR is the reflectance value of the near-infrared band; Red is the reflectance value of the red light band.
[0089] S2.3: Perform harmonic regression fitting on the normalized vegetation index and extract the statistical characteristics of the fitted normalized vegetation index, including: mean, variance, standard deviation, coefficient of variation and corrected variance;
[0090] In an optional embodiment, the time series of the normalized vegetation index is reconstructed by a harmonic regression fitting model, and the harmonic regression fitting model is:
[0091]
[0092] in, is the set of normalized vegetation indices after fitting reconstruction; n is the number of harmonics; t j is the time step, j = 1, 2, ..., N, N is the number of observations in the time series; p i and q i are the coefficients of the trigonometric functions in the Fourier series, fi is the corresponding frequency; x0 is a constant term.
[0093] In an optional implementation, statistical features of the fitted normalized difference vegetation index are extracted, and these statistical features are used to provide more abundant feature information for subsequent classifiers to obtain more accurate classification results.
[0094] For example, the time series of the reconstructed NDVI is fitted: The mean value of all time points can be expressed as The formula for extracting the mean is:
[0095]
[0096] Among them, MeanNDVI is the mean, represented by μ; num is the total number of observations in the time series.
[0097] The variance is used to measure the time series of the normalized difference vegetation index after fitting and reconstruction. The formula for extracting the variance of the volatility of the NDVI relative to its mean is:
[0098]
[0099] Among them, Variance is the variance, and σ is used 2 express.
[0100] The standard deviation is the square root of the variance, which is used to describe the fluctuation range. The formula for extracting the standard deviation is:
[0101]
[0102] Among them, Standard Deviation is the standard deviation, represented by σ.
[0103] Coefficient of Variation (CV) is used to describe the time series of the reconstructed normalized vegetation index after fitting. The formula for extracting the coefficient of variation is:
[0104]
[0105] Where CV is the standard deviation.
[0106] In the time series, the maximum and minimum values represent the peak and lowest points of vegetation coverage, respectively. The formulas for extracting the maximum and minimum values are:
[0107]
[0108] Among them, MaxNDVI is the maximum value, Indicates; MinNDVI is the minimum value, express.
[0109] In addition, as an alternative, other time series feature extraction methods can also be used to extract statistical features, such as wavelet transform, Fourier transform and dynamic time warping.
[0110] S2.4: Multi-source remote sensing data also includes: terrain feature information and radar image data information; multi-source feature integration is performed on multi-spectral band information, statistical features, terrain feature information and radar image data information to obtain an integrated result;
[0111] S2.5: Extract the integration result to obtain texture features.
[0112] In an optional embodiment, multi-source feature integration is performed on Sentinel-2 time series images, Sentinel-1 time series images, Landsat time series images, multispectral remote sensing bands, terrain features and radar image data in multi-source remote sensing data, and texture features are extracted based on the integration results.
[0113] For example, texture features are important characteristics of images, which represent the pattern of image grayscale changes in a certain form in space. Texture features are extracted through the Gray-Level Co-occurrence Matrix (GLCM), which is used to represent the grayscale joint probability distribution between two pixels with a specific spatial relationship. Based on the GEE platform, the Glcm Texture function is used, which calculates the GLCM texture index around each pixel in each band, so that each texture feature generates 18 texture feature bands, as shown in Table 2.
[0114] Table 2 Texture features and their meanings
[0115]
[0116]
[0117] Further, the steps of extracting texture features using GLCM include:
[0118] Convert multi-source remote sensing data into a grayscale image with U gray levels, ranging from 0 to U-1;
[0119] Construct the gray-level co-occurrence matrix G(x,y), set U×U as the size of the gray-level co-occurrence matrix G(x,y); (x,y) is a pixel pair with two gray levels; x is the gray level of the first pixel in the pixel pair; y is the gray level of the second pixel in the pixel pair; f is the distance between the pixel pairs; α is the direction of the pixel pair;
[0120] Normalize the gray-level co-occurrence matrix G(x,y) to obtain the normalized gray-level co-occurrence matrix G'(x,y).
[0121]
[0122] Where ∑∑G(x,y) is the sum of all elements in the normalized gray-level co-occurrence matrix G'(x,y);
[0123] The texture features are obtained according to the normalized gray-level co-occurrence matrix G'(x, y). The texture features include contrast, entropy, correlation, and average, and the formulas are as follows:
[0124]
[0125]
[0126] Among them, CONTRAST is contrast; ENT is entropy; CORR is correlation; SAVG is sum average; μ i is the mean of the row index; μ j is the mean of the column index; σ i is the standard deviation of the row index; σ j is the standard deviation of the column index; U s(k) It is the sum of the probabilities of pixels satisfying x+y=k in the gray-level co-occurrence matrix G(x,y).
[0127] S3: respectively obtain training samples and verification samples, and filter the training samples according to the verification samples to obtain a first sample set;
[0128] In an optional embodiment, S3 includes:
[0129] S3.1: Obtain training samples and validation samples respectively, where the validation samples are forest layer data;
[0130] Exemplarily, the training samples come from an online network system (https: / / survey.chinavegetation.cn / survey), which includes vegetation-related training samples, and records the attributes of vegetation groups or subgroups and their geographic coordinates in each sample. To ensure the accuracy of the samples, GlobeLan d30 data is used as a verification sample. It was developed by the National Geographic Information Center of China and covers the land cover in 2020. It covers 10 major land surface types including forests, grasslands, wetlands, water bodies, and artificial surfaces, and provides reliable forest boundary information for verifying the correctness of the training samples.
[0131] S3.2: Compare the training samples with the forest layer data, and select the training samples in the forest layer data as the reserved samples;
[0132] For example, the artificial forest samples and natural forest samples in the training samples are compared with the forest layer of the GlobeLand30 data. Through screening, the training samples in the GlobeLand30 forest layer are retained, and the samples not in the forest layer are considered to be inaccurate samples (wrong points) and are removed.
[0133] S3.3: Perform label reprojection on the retained samples to obtain the first sample set.
[0134] Exemplarily, after completing the sample screening, the labels of the retained samples are reprojected. Specifically, the labels of the natural forest samples are set to 0, and the labels of the artificial forest samples are set to 1.
[0135] S4: training a plurality of classifiers according to the first sample set, to obtain a plurality of first classifiers, and classifying the first sample set by the plurality of first classifiers, to obtain a plurality of first classification results;
[0136] In an optional embodiment, S4 includes:
[0137] S4.1: Select minority class samples and majority class samples from the first sample set as training sample sets respectively;
[0138] Among them, let the ratio of the number of minority classes to the number of majority classes be K. When 0<K≤2 / 3, it is considered that there is class imbalance in the samples in the sample set. 70% of the minority class samples are selected as the training set, and 30% of the minority class samples are selected as the validation set; the ratio of majority class samples as training samples varies, and generally the number of majority class samples that can obtain the maximum recall rate is selected. Correspondingly, after determining the number of majority class training samples, the number of majority class samples as validation samples is determined as follows: the number of majority class training samples: the number of majority class validation samples = 7:3.
[0139] S4.2: training a plurality of classifiers according to the training sample sets, the plurality of classifiers including: a random forest classifier, a decision tree classifier and a support vector machine classifier;
[0140] S4.3: adjusting the proportion of minority class samples in the training sample set, and retraining a plurality of classifiers according to the adjusted training sample set until the evaluation index is satisfied, thereby obtaining a plurality of first classifiers; wherein the evaluation index is used to evaluate the classification accuracy of the plurality of trained classifiers;
[0141] S4.4: Classify the first sample set according to a plurality of first classifiers respectively, and obtain corresponding first classification results.
[0142] For example, in order to deal with the problem of class imbalance, a smaller number of samples (minority samples), such as artificial forest samples, are selected from the first sample set as the benchmark, and its samples are divided into 70% training set and 30% validation set. The proportion of minority samples in the training sample set is adjusted, and continuous training is performed to improve the classifier's ability to identify minority samples, so as to achieve sample enhancement in subsequent steps based on its first classification results.
[0143] Specifically, the classifier with the maximum recall rate has been obtained by adjusting the parameters of the classifier during the training and classification process through evaluation indicators, including: confusion matrix results, overall precision and recall rate;
[0144] Confusion matrix results, including true positives TP, false positives FP, false negatives FN, and true negatives TN, are used to determine the classification accuracy of several classifiers;
[0145] The formula for overall accuracy is:
[0146]
[0147] Among them, Overall Accuracy is the overall accuracy.
[0148] The formula for recall is:
[0149]
[0150] Among them, Recall is the recall rate.
[0151] S5: performing sample enhancement on a minority of the first classification results to obtain a second sample set;
[0152] In an optional embodiment, if Figure 3 As shown, S5 includes:
[0153] S5.1: Calculate the overlap between a few of the first classification results and the forest layer data to obtain an overlap area, where the overlap area is:
[0154] A overlap =A s ∩A f ; (20)
[0155] Among them, A overlap A is the overlap area between the minority classification results and the forest layer data; s A is the area of minority classification results; f The area of the forest layer data.
[0156] S5.2: creating a buffer zone with the minority classification results located in the overlapped area as the center, generating new sample points in the buffer zone to achieve sample enhancement, and obtaining a second sample set based on the enhanced samples;
[0157] The formula for the buffer is:
[0158] B i ={(x,y)|(x i ) 2 +(yy i ) 2 ≤r 2}; (twenty one)
[0159] Where (x, y) is any point in the buffer zone; p i is the minority classification result, and its coordinates are (x i ,y i );B i is the minority classification result p i The buffer zone with center at and radius r.
[0160] The formula for the new sample point is:
[0161]
[0162] Among them, p new is the new sample point generated in the buffer; (x j ,y j ) is the coordinate of the new sample point generated in the buffer; n new The number of new sample points generated in the buffer.
[0163] The principle is: by comparing the minority classification results with the forest layer data, a map of the minority class distribution is obtained. Subsequently, the base layer for generating new sample points is obtained based on the minority class distribution map and the buffer zone. New sample points are generated in the base layer, that is, the enhanced sample points are all located in a reasonable forest coverage area.
[0164] S6: training a plurality of first classifiers according to the second sample set, and obtaining a plurality of corresponding second classifiers; classifying the preprocessed multi-source remote sensing data by the plurality of second classifiers, and obtaining a plurality of corresponding second classification results; and performing decision fusion on the plurality of second classification results to obtain a final classification result.
[0165] In an optional embodiment, if Figure 4 As shown, S6 includes:
[0166] S6.1: training a plurality of first classifiers according to the second sample set, and obtaining a plurality of second classifiers accordingly;
[0167] Exemplarily, the second sample set is used as input to train three classifiers, namely, a random forest classifier (RF), a decision tree classifier (DT) and a support vector machine classifier (SVM). Specifically, in the random forest classifier, a final classification result is generated by constructing multiple decision trees and taking the average or majority vote of their prediction results. Each decision tree is trained by different sample subsets and feature subsets, so the random forest classifier can reduce the overfitting problem of a single decision tree and improve the generalization ability of the model. In the decision tree classifier, a final classification result is generated by constructing multiple decision trees and taking the average or majority vote of their prediction results. Each decision tree is trained by different sample subsets and feature subsets, so the decision tree classifier can reduce the overfitting problem of a single decision tree and improve the generalization ability of the model. In the support vector machine classifier, by finding the best hyperplane, samples of different categories are separated, and the distance from the hyperplane to the closest sample point is maximized, which is particularly outstanding for classification problems in high-dimensional space.
[0168] S6.2: classifying the preprocessed multi-source remote sensing data according to a plurality of second classifiers, and obtaining a plurality of classification results accordingly;
[0169] S6.3: Decision fusion is performed on several second classification results to obtain a final classification result. The final classification result includes:
[0170]
[0171] in, is the final classification result; z is the pixel point; mode is the function for calculating the category with the most occurrences among the three classification results, i.e., the maximum voting method; and The following are the classification results of several second classifiers.
[0172] In addition to the random forest classifier, decision tree classifier, and support vector machine classifier, other machine learning algorithms or classifiers can be selected for classification, such as the extreme gradient boosting (XGBoost) method, deep learning models, and k-nearest neighbor classifiers. Decision fusion can also be performed based on other decision fusion strategies, such as weighted voting and adaptive enhancement methods.
[0173] The artificial forest multi-source remote sensing extraction method of the present invention is based on multi-source remote sensing data, performs sample enhancement according to the first classification result of the trained first classifier, and generates new sample points in the buffer zone of the minority class sample points, which not only effectively increases the number of minority class samples, but also ensures the reasonable distribution of the new sample points, improves the classifier's recognition ability for minority class samples, and solves the problem of class imbalance.
[0174] The artificial forest multi-source remote sensing extraction method of the present invention utilizes the time series characteristics of NDVI processing and extracts the trend and seasonal change characteristics in the time series through harmonic regression fitting, which not only enhances the classifier's ability to capture vegetation changes, but also improves the classifier's sensitivity to long-term changes and periodic changes, providing support for obtaining more accurate classification results.
[0175] The artificial forest multi-source remote sensing extraction method of the present invention trains three different classifiers, namely random forest, decision tree and support vector machine, and performs decision fusion to generate the final classification result. The fusion strategy can effectively integrate the advantages of different classifiers, overcome the bias problem that is prone to occur in a single classifier, and improve the comprehensive accuracy, stability and robustness of the final classification result.
[0176] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant is intended to cover non-exclusive inclusion, so that the article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the article or device including the elements. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The orientation or position relationship indicated by "up", "down", "left", "right", etc. is based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.
[0177] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.
Claims
1. A multi-source remote sensing extraction method for artificial forests, characterized in that: include: Acquire multi-source remote sensing data; Preprocessing the multi-source remote sensing data to obtain preprocessed multi-source remote sensing data; Acquire training samples and verification samples respectively, and filter the training samples according to the verification samples to obtain a first sample set; Training a plurality of classifiers according to the first sample set to obtain a plurality of first classifiers, and classifying the first sample set by the plurality of first classifiers to obtain a plurality of first classification results; Performing sample enhancement on a minority of the first classification results to obtain a second sample set; The several first classifiers are trained respectively according to the second sample set to obtain several corresponding second classifiers; the pre-processed multi-source remote sensing data are classified respectively by the several second classifiers to obtain several corresponding second classification results, and the several second classification results are decision-fused to obtain the final classification result.
2. The method for multi-source remote sensing extraction of artificial forests according to claim 1, characterized in that: Acquiring the multi-source remote sensing data includes: acquiring Sentinel-2 time series images, Sentinel-1 time series images, and Landsat time series images respectively.
3. The method for multi-source remote sensing extraction of artificial forests according to claim 2, characterized in that: Preprocessing the multi-source remote sensing data to obtain preprocessed multi-source remote sensing data includes: De-clouding the Sentinel-2 time series images, the Sentinel-1 time series images, and the Landsat time series images to obtain multispectral band information; Performing NDVI processing on the Landsat time series images after declouding to obtain a normalized vegetation index; Performing harmonic regression fitting on the normalized vegetation index, and extracting statistical characteristics of the fitted normalized vegetation index; wherein the statistical characteristics include: mean, variance, standard deviation, coefficient of variation and corrected variance; The multi-source remote sensing data also includes: terrain feature information and radar image data information; multi-source feature integration is performed on the multi-spectral band information, the statistical features, the terrain feature information and the radar image data information to obtain an integrated result; The integration result is extracted to obtain texture features.
4. The method for multi-source remote sensing extraction of artificial forests according to claim 3, characterized in that: The formula for the NDVI processing is: Among them, NDVI is the normalized vegetation index; NIR is the reflectance value of the near-infrared band; Red is the reflectance value of the red light band.
5. The method for multi-source remote sensing extraction of artificial forests according to claim 4, characterized in that: The model of the harmonic regression fitting is: in, is the set of normalized vegetation indices after fitting and reconstruction; n is the number of harmonics; t j is the time step, j = 1, 2, ..., N, N is the number of observations in the time series; p i and q i are the coefficients of the trigonometric functions in the Fourier series, f i is the corresponding frequency; x0 is a constant term.
6. The method for multi-source remote sensing extraction of artificial forests according to claim 1, characterized in that: Acquiring training samples and verification samples respectively, and screening the training samples according to the verification samples to obtain a first sample set, including: Obtain training samples and verification samples respectively, wherein the verification samples are forest layer data; Comparing the training samples with the forest layer data, and selecting the training samples in the forest layer data as reserved samples; Label reprojection is performed on the reserved samples to obtain a first sample set.
7. The method for multi-source remote sensing extraction of artificial forests according to claim 1, characterized in that: Several classifiers are trained according to the first sample set to obtain several first classifiers, and the first sample set is classified by the several first classifiers to obtain first classification results, including: Selecting minority class samples and majority class samples from the first sample set as training sample sets respectively; The plurality of classifiers are trained respectively according to the training sample sets; wherein the plurality of classifiers include: a random forest classifier, a decision tree classifier and a support vector machine classifier; Adjusting the proportion of minority class samples in the training sample set, retraining the plurality of classifiers according to the adjusted training sample set until an evaluation index is met, and obtaining a plurality of first classifiers; wherein the evaluation index is used to evaluate the classification accuracy of the plurality of classifiers after training; The first sample sets are classified respectively according to the plurality of first classifiers, and corresponding first classification results are obtained.
8. The method for multi-source remote sensing extraction of artificial forests according to claim 7, characterized in that: The evaluation indicators include: confusion matrix results, overall precision and recall; The confusion matrix result includes true positive examples TP, false positive examples FP, false negative examples FN and true negative examples TN, which are used to judge the classification accuracy of the several classifiers; The formula for the overall accuracy is: Among them, Overall Accuracy is the overall accuracy; The formula for the recall rate is: Among them, Recall is the recall rate.
9. The method for multi-source remote sensing extraction of artificial forests according to claim 6, characterized in that: Performing sample enhancement on a minority of the plurality of first classification results to obtain a second sample set includes: A minority of the first classification results are overlapped with the forest layer data to obtain an overlapped area, where the overlapped area is: A overlap =A s ∩A f ; Among them, A overlap is the overlapping area between the minority classification result and the forest layer data; s is the area of the minority classification results; f is the area of the forest layer data; Creating a buffer zone with the minority classification results located in the overlapped area as the center, generating new sample points in the buffer zone to achieve sample enhancement, and obtaining a second sample set according to the enhanced samples; The formula for the buffer zone is: B i ={(x,y)|(xx i ) 2 +(yy i ) 2 ≤r 2 }; Wherein, (x, y) is any point in the buffer zone; p i is the minority classification result, and its coordinates are (x i ,y i );B i is the minority classification result p i The buffer zone with the center being the circle and the radius being r; The formula for the new sample point is: Among them, p new is a new sample point generated in the buffer; (x j ,y j ) are the coordinates of the new sample points generated in the buffer; n new is the number of new sample points generated in the buffer.
10. The method for multi-source remote sensing extraction of artificial forests according to claim 1, characterized in that: The training of the plurality of first classifiers according to the second sample set to obtain a plurality of second classifiers includes: Training the plurality of first classifiers respectively according to the second sample set to obtain the plurality of second classifiers accordingly; Classifying the pre-processed multi-source remote sensing data according to the plurality of second classifiers, and obtaining a plurality of second classification results accordingly; The plurality of second classification results are subjected to decision fusion to obtain a final classification result, wherein the final classification result includes: in, is the final classification result; z is the pixel point; mode is the function for calculating the category with the most occurrences among the three classification results; and The classification results of the several second classifiers are shown in order.
Citation Information
Patent Citations
Multi-source remote sensing image classification method based on two-way attention fusion neural network
CN109993220A
Large-scale soybean planting region extraction algorithm based on Google Earth Engine
CN113657158A
Multi-temporal active and passive remote sensing random forest crop identification method and system
CN115035413A
Remote sensing classification method fusing vegetation life cycle characteristics
CN117689959A
Building structure automatic classification method combining SMOTE-ENN resampling technology and random forest
CN118427726A
Cited By
Flood water body identification method, system and device and storage medium
CN120298790A
Rice information extraction method and equipment based on multi-source remote sensing and machine learning
CN121661528A
