Method and device for determining forest disturbance type, and electronic equipment

Abstract

The invention discloses a method and device for determining a forest disturbance type and electronic equipment, and the method comprises the steps: obtaining forest disturbance data from a land coverage data set; performing feature analysis on the forest disturbance data to obtain forest disturbance data under each classification; taking the forest disturbance data under each classification as basic data, and performing multiple iterations in a mode of mutual conductance of a random forest classifier and a support vector machine classifier to perform sample data volume amplification and classifier training to obtain the random forest classifier and the support vector machine classifier which meet preset conditions; and predicting to-be-predicted forest disturbance data based on the random forest classifier and the support vector machine classifier which meet preset conditions to obtain a forest disturbance type. According to the scheme for determining the forest disturbance type, time consumed by sample marking and labor cost can be saved, and the forest disturbance type can be accurately identified.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing monitoring of ecological environment in a small sample neighborhood, and in particular to a method and device for determining forest disturbance types, and electronic equipment. Background Art

[0002] Disturbance is an important part of the dynamic change process of forest ecosystems. It is heterogeneous in time and space and is a relatively discrete event in time. It can change the composition structure of forests, weaken the ecological service capacity, and reduce the availability of forest resources. Disturbance not only affects the evolution of forest ecosystems themselves, but also has a strong impact on forest carbon storage and carbon sinks. In particular, subtropical evergreen forests are important global carbon sinks. However, due to frequent natural disasters (such as fires) and human activities (such as reclamation and logging), subtropical forests have become a hot spot for forest disturbance research. In order to better quantify forest loss and regional and global carbon dynamics and maintain the sustainability of forest ecosystems, it is essential to understand the characteristics and impacts of different disturbance types in subtropical evergreen forests.

[0003] At present, the identification methods of forest disturbance types can be summarized into two categories, namely direct method and two-stage method. The direct method is to use the time-spectral measurement indicators obtained from remote sensing time series data and the prior knowledge of disturbance types to simultaneously detect and attribute disturbances at the pixel scale with the help of machine learning algorithms (such as random forests, decision trees, etc.). The two-stage method is divided into two steps. First, the disturbance event is determined based on the change detection algorithm at the pixel scale. Then, after the disturbance pixels are merged into disturbance patches, the spatial spectrum characteristics of the disturbance event are extracted. Finally, the disturbance event is attributed with the help of machine learning algorithms (such as random forests, deep learning algorithms, etc.). The essence of the above two methods is to use machine learning for classification. However, this requires a large amount of sample data (prior knowledge) to train the model. Many studies manually interpret a large amount of sample data through various historical records, high-resolution images, field surveys, etc. when obtaining sample data. In order to improve the generalization performance of the classification model, at least 200 labeled samples are required for each type of disturbance. The labeling cost is very high and takes a lot of time. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method and device for determining the type of forest disturbance, and an electronic device, which can solve the above-mentioned problems existing in the prior art.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In one aspect of the present invention, a method for determining a forest disturbance type is provided, comprising:

[0007] Obtaining forest disturbance data from a land cover dataset; wherein the disturbance data includes: disturbance time and disturbance occurrence location;

[0008] Performing feature analysis on the forest disturbance data to obtain forest disturbance data under each classification;

[0009] Taking the forest disturbance data under each classification as basic data, the random forest classifier and the support vector machine classifier are iteratively amplified for sample data and trained for classifiers multiple times to obtain random forest classifiers and support vector machine classifiers that meet preset conditions;

[0010] The forest disturbance data to be predicted is predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type.

[0011] Optionally, the step of obtaining forest disturbance data from a land cover dataset comprises:

[0012] The initial land cover dataset was subjected to forest masking to obtain forest-related data;

[0013] The forest-related data is detected using a continuous change detection and classification algorithm CCDC to obtain a detection result, wherein the detection result includes: a disturbance caused by a forest changing into other land types and a disturbance caused by other land types changing into a forest;

[0014] Based on the land cover dataset, disturbance pixels of the forest change to other land types are filtered to obtain forest disturbance data.

[0015] Optionally, the step of performing feature analysis on the forest disturbance data to obtain forest disturbance data under each category includes:

[0016] Analyzing the forest disturbance data based on the temporal characteristics, spatial characteristics, spectral characteristics and topographic characteristics of the subtropical evergreen forest disturbance classification to obtain the classification corresponding to each forest disturbance;

[0017] The forest disturbance data of the same classification are divided into the same classification to obtain the forest disturbance data under each classification.

[0018] Optionally, the steps of using the forest disturbance data under each classification as basic data, iterating multiple times by mutual induction of a random forest classifier and a support vector machine classifier to amplify the sample data volume and train the classifier to obtain a random forest classifier and a support vector machine classifier that meet preset conditions include:

[0019] For each classification of forest disturbance data, a preset number of forest disturbance data are taken as sample data;

[0020] Divide the sample data of each category into training samples and test samples according to a preset ratio;

[0021] The training samples are used to train the random forest classifier and the support vector machine classifier respectively;

[0022] Predicting the test samples corresponding to each classification based on the trained random forest classifier and support vector machine classifier, and determining the test accuracy of the random forest classifier and support vector machine classifier;

[0023] When the test accuracy does not meet the preset conditions, classifying the unlabeled forest disturbance data according to the trained random forest classifier and support vector machine classifier to obtain a first classification result and a second classification result;

[0024] Determining training sample increment samples of the random forest classifier and the support vector machine classifier according to the first classification result and the second classification result;

[0025] The random forest classifier is trained using the incremental training samples corresponding to the random forest classifier, and the support vector machine classifier is trained using the incremental training samples corresponding to the support vector machine classifier, thereby completing a mutual training of the random forest classifier and the support vector machine classifier.

[0026] Optionally, the step of determining the training sample increment samples of the random forest classifier and the support vector machine classifier according to the first classification result and the second classification result includes:

[0027] Comparing the first classification result with the second classification result, and determining the classification result with a difference as pseudo-label data;

[0028] Filtering out first high-confidence pseudo-label data from the pseudo-label data included in the first classification result, and using the first high-confidence pseudo-label data as a training sample increment sample of the support vector machine classifier;

[0029] Second high-confidence pseudo-label data are screened out from the pseudo-label data included in the second classification result, and the second high-confidence pseudo-label data are used as training sample increment samples of the random forest classifier.

[0030] Optionally, the step of selecting first high-confidence pseudo-label data from the pseudo-label data included in the first classification result includes:

[0031] From the pseudo-label data included in the first classification result, first pseudo-label data having a classification probability value greater than a first preset probability threshold is selected;

[0032] A confidence learning algorithm is used to filter out erroneous pseudo-label data in the first pseudo-label data to obtain first high-confidence pseudo-label data.

[0033] Optionally, the step of using the first high-confidence pseudo-label data as a training sample increment sample of the support vector machine classifier includes:

[0034] For each category included in the first classification result, determine first high-confidence pseudo-label data corresponding to each category;

[0035] Perform inter-class data balancing processing on the first high-confidence pseudo-label data corresponding to each classification;

[0036] The first high-confidence pseudo-label data after inter-class balancing processing is used as a training sample increment sample of the support vector machine classifier.

[0037] In another aspect of the present invention, there is provided a device for determining the type of forest disturbance, wherein:

[0038] An acquisition module is used to acquire forest disturbance data from a land cover data set; wherein the disturbance data includes: disturbance time and disturbance occurrence location;

[0039] A classification module, used for performing feature analysis on the forest disturbance data to obtain forest disturbance data under each classification;

[0040] A mutual guidance training module is used to use the forest disturbance data under each classification as basic data, and adopt a random forest classifier and a support vector machine classifier to iteratively amplify the sample data and train the classifier multiple times in a mutual guidance manner to obtain a random forest classifier and a support vector machine classifier that meet preset conditions;

[0041] The prediction module is used to predict the forest disturbance data to be predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type.

[0042] Optionally, the acquisition module includes:

[0043] The first submodule is used to perform forest masking on the initial land cover dataset to obtain forest-related data;

[0044] The second submodule is used to detect the forest-related data using a continuous change detection and classification algorithm CCDC to obtain a detection result, wherein the detection result includes: a disturbance in which a forest changes into other land types and a disturbance in which other land types change into a forest;

[0045] The third submodule is used to filter the disturbance pixels of the forest change into other land types based on the land cover dataset to obtain forest disturbance data.

[0046] In another aspect of the implementation of the present invention, there is also provided an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to implement any of the above-mentioned methods for determining the type of forest disturbance when executing the program stored in the memory.

[0047] The embodiment of the present invention provides a scheme for determining the type of forest disturbance, which obtains forest disturbance data from a land cover data set; performs feature analysis on the forest disturbance data to obtain forest disturbance data under each classification; uses the forest disturbance data under each classification as basic data, and adopts a random forest classifier and a support vector machine classifier to iteratively amplify the sample data volume and train the classifier multiple times in a mutual induction manner to obtain a random forest classifier and a support vector machine classifier that meet preset conditions; predicts the forest disturbance data to be predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type. Through the scheme for determining the type of forest disturbance provided by the embodiment of the present invention, the model is trained by combining small sample learning and the mutual induction learning method of the two base classifiers of the random forest and the support vector machine. The user only needs to manually mark a small number of samples, and then the sample data volume is amplified by iterative mutual induction, which can save the time and labor cost of sample marking. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flowchart showing the steps of a method for determining a forest disturbance type according to an embodiment of the present application;

[0049] Figure 2 is a flow chart showing another method for determining forest disturbance type according to an embodiment of the present application;

[0050] Figure 3 is a diagram showing a classifier mutual guidance framework based on confidence learning according to an embodiment of the present application;

[0051] Figure 4 It is a structural block diagram of a device for determining forest disturbance type according to an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0053] Therefore, in order to reduce the annotation cost of sample data volume and solve the problem that the change area of ​​a certain type of change type accounts for a small proportion compared with other types of change, resulting in a small amount of sample data for this type of change, it is of great significance to study the forest disturbance type identification method under small sample conditions. Classification under small sample conditions is also called small sample learning. It is currently applied in the field of hyperspectral and belongs to semi-supervised learning. Semi-supervised learning uses a small amount of sample data to train the model, and then uses the trained model to predict a large amount of unlabeled data. The prediction result is called a pseudo-label. High-quality pseudo-labels are selected from the prediction results and added to a small amount of labeled data to expand the sample data, thereby reducing manual participation from a mechanism perspective. It can be seen that how to select high-quality pseudo-labels to add to the training set is the focus of semi-supervised learning. Existing studies have mainly expanded sample data from two aspects. One is to select high-quality pseudo-labels based on similarity metrics such as cosine distance within a certain neighborhood. This method has the problem that the number of selected pseudo-labels is limited and noise cannot be effectively solved; the other is the uncertainty and confidence selection based on probability. This method can integrate multiple classification models to expand the training set, solving the problem of pseudo-label number limitation, but it uses its own model to iteratively predict, and the existence of noisy pseudo-labels is prone to error accumulation. To this end, some scholars have proposed a mutual guidance framework. The mutual guidance framework adds the high-confidence pseudo-label data obtained by each of the two classifiers to the other party's small sample set to eliminate the error accumulation problem that may be caused by noise iterative training of the same classification model. However, the noise problem still exists in the mutual guidance framework, and the selection of high-confidence pseudo-labels in each iteration is a fixed number, without considering the balance of the number of different types of pseudo-labels, and there is also a certain data waste problem. Based on this, the present invention designs a scheme for determining the type of forest disturbance, which is a forest disturbance attribution based on a confidence learning mutual guidance framework under small sample conditions.

[0054] The following is a detailed description of the scheme for determining the forest disturbance type provided in the embodiment of the present application through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0055] As attached Figure 1 As shown, the method for determining the forest disturbance type in the embodiment of the present application includes the following steps:

[0056] Step 101: Get forest disturbance data from the land cover dataset.

[0057] The disturbance data includes: the disturbance time and the disturbance occurrence location.

[0058] Land cover dataset: Forest masks are extracted using existing land cover classification data. The data comes from Wuhan University, which builds spatiotemporal features based on all available Landsat data on GEE, combines random forest classifier classification, and proposes a post-processing method including spatiotemporal filtering and logical reasoning to further improve the spatiotemporal consistency of CLCD. The annual China Land Cover Dataset (CLCD) is produced, which contains annual land cover information in China from 1985 and 1990 to 2020. Finally, the classification accuracy is verified based on 5463 visual interpretation samples, and the overall accuracy of CLCD can reach 80%.

[0059] Continuous Change Detection and Classification (CCDC): It is a disturbance detection algorithm for all types of objects based on time series data proposed by Zhu and Woodcock. It can identify all three types of changes: seasonal changes, gradual changes, and sudden changes, mainly for identifying sudden changes. The principle of change detection is: use ordinary least squares (OLS) to model the seasonal changes and gradual changes in pixel values ​​of each band in the image over time. Then the difference between the predicted modeled pixel value and the true pixel value will be calculated. When the difference between the values ​​is greater than three times the root mean square error (RMSE), the pixel will be marked as a possible land cover change. Then, based on the number of consecutive observations, it is evaluated whether the potential land cover change is a real change.

[0060] In an embodiment of the present application, forest disturbance data is obtained based on a land cover dataset and CCDC. In an optional embodiment, a method of obtaining forest disturbance data from a land cover dataset may include the following sub-steps:

[0061] Sub-step 1: Perform forest masking on the initial land cover dataset to obtain forest-related data;

[0062] Sub-step 2: using a continuous change detection and classification algorithm CCDC to detect the forest-related data and obtain a detection result;

[0063] The detection results include: disturbances caused by changes in forests to other land types and disturbances caused by changes in other land types to forests.

[0064] Sub-step 3: Based on the land cover dataset, the disturbance pixels of forest changes to other land types are filtered to obtain forest disturbance data.

[0065] Step 102: Perform feature analysis on the forest disturbance data to obtain forest disturbance data under each category.

[0066] In this step, the spatiotemporal spectrum and four topographic features suitable for the classification of subtropical evergreen forest disturbance are found to distinguish different types of forest disturbance.

[0067] An optional method of performing feature analysis on forest disturbance data to obtain forest disturbance data under each category may be as follows:

[0068] Based on the temporal characteristics, spatial characteristics, spectral characteristics and terrain characteristics of the subtropical evergreen forest disturbance classification, the forest disturbance data are analyzed to obtain the classification corresponding to each forest disturbance; the forest disturbance data of the same classification are divided into the same classification to obtain the forest disturbance data under each classification.

[0069] Temporal characteristics: Temporal characteristics are the disturbance amplitude, duration, and rate of disturbance that characterize different types of forest disturbances;

[0070] Spatial characteristics: Spatial characteristics characterize the disturbance area and texture differences of different forest disturbance types;

[0071] Spectral characteristics: Spectral characteristics characterize the differences in spectral values ​​or vegetation index values ​​before and after disturbance for different forest disturbance types;

[0072] Topographic characteristics: Topographic characteristics characterize the differences in spatial locations of different forest disturbance types.

[0073] Step 103: Using the forest disturbance data under each classification as basic data, the random forest classifier and the support vector machine classifier are iteratively used to amplify the sample data volume and train the classifier multiple times to obtain the random forest classifier and the support vector machine classifier that meet the preset conditions.

[0074] Random Forest Classifier, Random Forest algorithm (RF) uses a random method to build a forest, which is composed of several unrelated decision trees. After obtaining the forest, when a new input sample enters, each decision tree in the forest is allowed to distinguish the category of the sample (for the classification algorithm), and finally vote on the final category of the sample.

[0075] Support Vector Machine Classifier,Support Vector Machine (SVM) is an efficient supervised learning algorithm. Its goal is to maximize the category margin by finding the optimal classification interface equation, thereby achieving accurate classification.

[0076] The mutual guidance based on RF and SVM is used to provide pseudo labels to each other, guide each other's learning, and avoid the accumulation of errors. RF and SVM use a small amount of sample data to preliminarily classify a large amount of unlabeled data for the first time, and obtain two different classification results, namely pseudo labels. Select high-confidence pseudo labels from the pseudo labels obtained from RF and add them to the small amount of sample data of SVM to expand the sample set of SVM. At the same time, select high-confidence pseudo labels from the pseudo labels obtained from SVM and add them to the small amount of sample data of RF to expand the sample set of RF, thus realizing the mutual guidance of RF and SVM.

[0077] In the embodiment of the present application, mutual guidance based on random forest and support vector machine classifiers is used to reduce the cost of manually annotating samples, realize mutual teaching of classification under small sample conditions, and provide optional high-confidence pseudo-sample data; high-confidence pseudo-labels are selected, high-confidence pseudo-label data are screened based on classification probability, and confidence learning is introduced to remove erroneous pseudo-labels in high-confidence pseudo-label data, balance the number of pseudo-labels in each category, and use it to expand the sample data volume, which can achieve high-precision forest disturbance attribution under small sample conditions.

[0078] Step 104: Predict the forest disturbance data to be predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type.

[0079] The preset conditions can be flexibly set by those skilled in the art, for example: set to the prediction accuracy after two iterations of training is less than a preset threshold; for example: it can be set to the prediction accuracy of the classifier reaches a preset accuracy after one iteration of training; for another example: the number of iterations reaches a preset number of iterations, etc. The preset conditions are not specifically limited in the embodiments of the present application.

[0080] The method for determining the type of forest disturbance provided in the embodiment of the present application obtains forest disturbance data from a land cover data set; performs feature analysis on the forest disturbance data to obtain forest disturbance data under each classification; uses the forest disturbance data under each classification as the basic data, and uses a random forest classifier and a support vector machine classifier to iteratively amplify the sample data volume and train the classifier multiple times to obtain a random forest classifier and a support vector machine classifier that meet preset conditions; predicts the forest disturbance data to be predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type. Through the scheme for determining the type of forest disturbance provided by the embodiment of the present invention, the model is trained by combining small sample learning and the mutual guidance learning method of the two base classifiers of random forest and support vector machine. The user only needs to manually mark a small number of samples, and then the sample data volume is amplified by iterative mutual guidance, which can save the time and labor cost of sample marking.

[0081] In an optional embodiment, the forest disturbance data under each classification is used as the basic data, and the random forest classifier and the support vector machine classifier are used to iterate multiple times to expand the sample data volume and train the classifier. The method of obtaining the random forest classifier and the support vector machine classifier that meet the preset conditions may include the following sub-steps:

[0082] Sub-step 1: For each classification of forest disturbance data, a preset number of forest disturbance data are taken as sample data.

[0083] Sub-step 2: Divide the sample data of each category into training samples and test samples according to a preset ratio.

[0084] An exemplary method of acquiring and dividing sample data is as follows: take one hundred sample data for each classification and randomly divide them into training samples and test samples according to a ratio of 7:3. It should be noted that the preset number and preset ratio can be flexibly set by those skilled in the art, and there is no specific limitation on this in the embodiments of the present application. For example, the preset ratio can be set to 8:2, 9:1, etc.

[0085] Sub-step 3: Use the training samples to train the random forest classifier and the support vector machine classifier respectively.

[0086] Sub-step 4: Predict the test samples corresponding to each classification based on the trained random forest classifier and support vector machine classifier, and determine the test accuracy of the random forest classifier and the support vector machine classifier.

[0087] When the test accuracy meets the preset conditions, the iterative training can be stopped to determine that the random forest classifier and the support vector machine classifier have been trained.

[0088] Sub-step 5: When the test accuracy does not meet the preset conditions, the unlabeled forest disturbance data is classified according to the trained random forest classifier and support vector machine classifier to obtain the first classification result and the second classification result.

[0089] Sub-step 6: Determine the training sample increment samples of the random forest classifier and the support vector machine classifier based on the first classification result and the second classification result.

[0090] More specifically, based on the first classification result and the second classification result, the method for determining the incremental samples of training samples for the random forest classifier and the support vector machine classifier can be: comparing the first classification result and the second classification result, and determining the classification results with differences as pseudo-label data (i.e., pseudo-labels); filtering out first high-confidence pseudo-label data from the pseudo-label data contained in the first classification result, and using the first high-confidence pseudo-label data as the incremental samples of training samples for the support vector machine classifier; filtering out second high-confidence pseudo-label data from the pseudo-label data contained in the second classification result, and using the second high-confidence pseudo-label data as the incremental samples of training samples for the random forest classifier.

[0091] An exemplary method for filtering out first high-confidence pseudo-label data from the pseudo-label data contained in the first classification result can be: filtering out first pseudo-label data whose classification probability value is greater than a first preset probability threshold from the pseudo-label data contained in the first classification result; using a confidence learning algorithm to filter out erroneous pseudo-label data in the first pseudo-label data to obtain first high-confidence pseudo-label data.

[0092] The method of selecting the second high-confidence pseudo-label data from the pseudo-label data included in the second classification result is similar to the above method and will not be repeated here.

[0093] In the actual implementation process, the method of selecting high-confidence pseudo labels may include the following steps:

[0094] First, the initial screening of pseudo labels is used to filter out pseudo labels with low probability values ​​and obtain pseudo label data with high confidence. Second, the filtering of wrong pseudo labels is used to filter out wrong pseudo labels in pseudo label data with high confidence.

[0095] Confidence learning is an emerging, principled framework to identify label errors, characterize label noise, and apply it to noisy learning. Confidence learning can directly estimate the joint distribution of noisy labels and true labels without making assumptions about random uniform label noise and hyperparameters. It only needs to use cross-validation to obtain out-of-sample prediction probabilities. Applying confidence learning to the multi-classifier mutual guidance framework can automatically find and filter samples that are incorrectly labeled by the base classifier.

[0096] Confidence learning to filter out erroneous samples is divided into two steps. The first step is to estimate the joint distribution of noise labels and true labels (count). The second step is to find and filter out erroneous samples (clean).

[0097] In the first step, the noise labels are defined as That is, samples that have been labeled by the base classifier but may contain errors; the true label is defined as y *, the true label is estimated by cross-validation. In addition, the total number of samples is defined as n and the total number of categories is defined as m.

[0098] To estimate the joint distribution, four steps are required:

[0099] a) Cross Validation

[0100] First, we need to calculate the probability P[i][j] of the i-th sample in the j-th category by cross-validating the data set, and then calculate the average probability t[j] of each base classifier in the j-th category as the confidence threshold; finally, for sample i, calculate its true label y * is the maximum probability argmaxP[i][j] among the j categories, and P[i][j]>t[j].

[0101]

[0102] Among them, θ is the model.

[0103] b) Calculate the count matrix (Similar to a confusion matrix)

[0104]

[0105] in,

[0106]

[0107] Where l∈[m];

[0108] c) Calibration count matrix

[0109] The purpose is to make the total count the same as the total number of samples labeled by the base classifier. The formula is as follows:

[0110]

[0111] in, Pseudo label The total number of samples.

[0112] d) Estimating noisy labels and the true label y * The joint distribution of

[0113]

[0114] The joint distribution can fully reflect the distribution of noisy (wrong) labels and true labels in the real world. As the data scale increases, this estimation method is closer to the true distribution.

[0115] After obtaining the joint distribution of noise labels and true labels, the erroneous samples can be filtered out.

[0116] An exemplary method of using the first high-confidence pseudo-label data as an incremental sample of training samples for a support vector machine classifier can be as follows: for each category included in the first classification result, determine the first high-confidence pseudo-label data corresponding to each category; perform inter-class data balancing processing on the first high-confidence pseudo-label data corresponding to each category; and use the first high-confidence pseudo-label data after inter-class balancing processing as an incremental sample of training samples for a support vector machine classifier.

[0117] In a similar way, the second highest confidence pseudo-label data can be used as the incremental training sample of the random forest classifier. The balance between the number of classes can eliminate the problem of low generalization performance of the classification model caused by the imbalance between the number of classes.

[0118] Sub-step 7: Use the incremental training samples corresponding to the random forest classifier to train the random forest classifier, and use the incremental training samples corresponding to the support vector machine classifier to train the support vector machine classifier, thereby completing a mutual training of the random forest classifier and the support vector machine classifier.

[0119] In the actual implementation process, sub-steps 4 to 7 can be repeatedly executed to perform multiple iterative mutual training on the random forest classifier and the support vector machine classifier until the two classifiers meet the preset conditions and the training is stopped.

[0120] Combine the following Figure 2 , Figure 3 The method for determining the forest disturbance type according to an embodiment of the present application is described with a specific example.

[0121] like Figure 2 As shown, the method for determining the type of forest disturbance in an embodiment of the present application mainly includes the following parts: data preprocessing, labeling of a small number of samples, construction and selection of attribution features, iterative mutual training of random forest classifiers and support vector machine classifiers based on a small number of samples, and determination of forest disturbance attribution.

[0122] The specific details of disturbance data acquisition, i.e. data preprocessing, are as follows:

[0123] In this example, CCDC is used to obtain forest disturbance pixels for subsequent disturbance attribution. Forest disturbance events are detected using CCDC in GEE (Google Earth Engine). First, all available Landsat data with cloud cover less than 80% are used, covering the period from 1986 to 2022. Ideally, the earliest forest is considered to be the most original forest range, so the forest mask is extracted using the CLCD data of 1985 and applied to all Landsat data. Secondly, in order to detect forest disturbances as much as possible, the algorithm's input parameters minObservations and chiSquareProbability are set to 4 and 0.9, respectively, and multiple disturbance data are detected. Finally, through visual interpretation, 250 change samples and 250 unchanged samples (sample points conform to the normal distribution) are produced for accuracy verification, with an overall accuracy of 90.4%.

[0124] Disturbance attribution is only for the type of forest to other land types. Although the forest mask is used to ensure the disturbance of forest, CCDC can detect disturbances of all land types, so after the second forest disturbance, there are still other land types that are converted to forest. In order to filter out this type, the time of each disturbance was checked and it was found that there was almost only one disturbance in a year. Therefore, the annual CLCD data was used to filter the pixels that changed from other land types to forests, and the disturbed pixels of forest to other land types were obtained.

[0125] Feature extraction is attribution feature construction and selection:

[0126] The forest disturbance types identified in this study include fire, logging, road construction and reclamation. Summarizing the features used in existing literature, analyzing the differences in time, space, spectrum and terrain of different disturbance types, the following characteristics are obtained.

[0127] Table 1 shows the characteristics of forest disturbance attribution

[0128]

[0129]

[0130] Iterative mutual training of random forest classifier and support vector machine classifier based on a small number of samples:

[0131] The framework of classifier mutual guidance based on confidence learning is shown in the figure Figure 3 First, the sample data of each category, i.e., the number of classifications is about 100, is randomly divided into training samples and test samples according to a ratio of 7:3.

[0132] Secondly, a small amount of sample data is used to train the RF and SVM models, and then the trained models are used to perform preliminary classification predictions on a large amount of unlabeled data, respectively, to obtain two different classification results, namely pseudo labels; at the same time, the classification results are verified by using test samples. Next, high-confidence pseudo labels are selected from the pseudo labels obtained by RF and added to a small amount of sample data of SVM to expand the sample set of SVM. At the same time, high-confidence pseudo labels are selected from the pseudo labels obtained by SVM and added to a small amount of sample data of RF to expand the sample set of RF. Then, the expanded sample set is used to train RF and SVM, and then the trained models are used to perform classification predictions on the updated unlabeled data to achieve mutual guidance between the RF and SVM models; at the same time, the classification results are verified by using test samples. Iterate in this way, and use the size of the verification accuracy and the difference between the two verification accuracies to judge the generalization performance and stability of the model. After stopping the classification, if the classification results of the two classifiers are consistent, the data is assigned to this category. If the two are inconsistent, the category with a large probability is selected.

[0133] The additional selection of high-confidence pseudo-labels includes:

[0134] First, the pseudo labels are preliminarily screened to obtain the probabilities Prob_SVM and Prob_RF during SVM and RF classification, and the empirical thresholds are set to filter out pseudo labels with low probability values. They are:

[0135] Pseudo_label SVM =Prob_SVM>=0.9;

[0136] Pseudo_label RF =Prob_RF>=0.7;

[0137] Filtering of incorrect pseudo labels:

[0138] Secondly, confidence learning is used to filter out incorrect pseudo labels. The filtering rules are as follows:

[0139] Select Sample filtering is to select samples whose subscript j corresponding to the maximum probability of P[i][j] is inconsistent with the pseudo label.

[0140] Balance of number between classes:

[0141] In the pseudo-label results obtained after filtering the wrong pseudo-labels, the number of pseudo-labels of different perturbation types is unbalanced. Inputting them into the classifier will cause the classifier to learn towards the category with a large number, resulting in low classification accuracy. Therefore, it is necessary to balance the number of classes. The rule for balancing the number of classes is set as:

[0142] Label_numMin=min(label_numi)

[0143] Among them, i∈[1, 2, 3, 4], represents different perturbation types;

[0144] If Label_numMin is less than 50, then

[0145] labeli_num = 50;

[0146] If Label_numMin is greater than or equal to 50 and less than 100, then

[0147] labeli_num=Label_numMin / / 2;

[0148] If Label_numMin is greater than or equal to 100 and less than 150, then

[0149] labeli_num=Label_numMin / / 3;

[0150] If Label_numMin is greater than or equal to 150 and less than 200, then

[0151] labeli_num=Label_numMin / / 4;

[0152] If Label_numMin is greater than or equal to 200 and less than 250, then

[0153] labeli_num=Label_numMin / / 5;

[0154] If Label_numMin is greater than or equal to 250 and less than 300, then

[0155] labeli_num=Label_numMin / / 6;

[0156] If Label_numMin is greater than or equal to 300, then

[0157] labeli_num = 50;

[0158] The method for determining the type of forest disturbance provided in this specific example explores the application of small samples in the attribution of disturbance in subtropical evergreen forests. Small sample learning is often used in the classification of high-resolution hyperspectral data, and has not been applied to the attribution of disturbances in evergreen forests based on medium resolution and long time series. At the same time, combined with confidence learning, it solves the noise problem and sample size balance problem in small sample learning. In addition, in this example, the mutual guidance of the two base classifiers, random forest and support vector machine, is used to perform classifier learning, i.e. training, which can improve the classification accuracy of the classifier.

[0159] Figure 4A structural block diagram of a device for determining forest disturbance types according to an embodiment of the present application.

[0160] The device for determining the forest disturbance type provided in the embodiment of the present application includes the following functional modules:

[0161] The acquisition module 401 is used to acquire forest disturbance data from the land cover data set; wherein the disturbance data includes: disturbance time and disturbance occurrence location;

[0162] A classification module 402 is used to perform feature analysis on the forest disturbance data to obtain forest disturbance data under each classification;

[0163] The mutual guidance training module 403 is used to use the forest disturbance data under each classification as basic data, and adopt the random forest classifier and the support vector machine classifier to iteratively amplify the sample data and train the classifier multiple times in a mutual guidance manner to obtain the random forest classifier and the support vector machine classifier that meet the preset conditions;

[0164] The prediction module 404 is used to predict the forest disturbance data to be predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type.

[0165] Optionally, the acquisition module includes:

[0166] The first submodule is used to perform forest masking on the initial land cover dataset to obtain forest-related data;

[0167] The second submodule is used to detect the forest-related data using a continuous change detection and classification algorithm CCDC to obtain a detection result, wherein the detection result includes: a disturbance in which a forest changes into other land types and a disturbance in which other land types change into a forest;

[0168] The third submodule is used to filter the disturbance pixels of the forest change into other land types based on the land cover dataset to obtain forest disturbance data.

[0169] Optionally, the classification module includes:

[0170] The fourth submodule is used to analyze the forest disturbance data based on the temporal characteristics, spatial characteristics, spectral characteristics and topographic characteristics of the subtropical evergreen forest disturbance classification to obtain the classification corresponding to each forest disturbance;

[0171] The fifth submodule is used to classify the forest disturbance data of the same category into the same category to obtain the forest disturbance data under each category.

[0172] Optionally, the mutual guidance training module includes:

[0173] A sixth submodule is used for taking a preset amount of forest disturbance data as sample data for each classification of forest disturbance data;

[0174] The seventh submodule is used to divide the sample data of each classification into training samples and test samples according to a preset ratio;

[0175] An eighth submodule is used to respectively train a random forest classifier and a support vector machine classifier using training samples;

[0176] A ninth submodule is used to predict the test samples corresponding to each classification according to the trained random forest classifier and support vector machine classifier, and determine the test accuracy of the random forest classifier and the support vector machine classifier;

[0177] A tenth submodule is used for classifying the unlabeled forest disturbance data according to the trained random forest classifier and support vector machine classifier to obtain a first classification result and a second classification result when the test accuracy does not meet the preset condition;

[0178] An eleventh submodule is used to determine the training sample increment samples of the random forest classifier and the support vector machine classifier according to the first classification result and the second classification result;

[0179] The twelfth submodule is used to train the random forest classifier using the incremental training samples corresponding to the random forest classifier, and to train the support vector machine classifier using the incremental training samples corresponding to the support vector machine classifier, so as to complete a mutual training of the random forest classifier and the support vector machine classifier.

[0180] Optionally, the eleventh submodule is specifically used for:

[0181] Comparing the first classification result with the second classification result, and determining the classification result with a difference as pseudo-label data;

[0182] Filtering out first high-confidence pseudo-label data from the pseudo-label data included in the first classification result, and using the first high-confidence pseudo-label data as a training sample increment sample of the support vector machine classifier;

[0183] Second high-confidence pseudo-label data are screened out from the pseudo-label data included in the second classification result, and the second high-confidence pseudo-label data are used as training sample increment samples of the random forest classifier.

[0184] Optionally, when the eleventh submodule selects the first high-confidence pseudo-label data from the pseudo-label data included in the first classification result, it is specifically used to:

[0185] From the pseudo-label data included in the first classification result, first pseudo-label data having a classification probability value greater than a first preset probability threshold is selected;

[0186] A confidence learning algorithm is used to filter out erroneous pseudo-label data in the first pseudo-label data to obtain first high-confidence pseudo-label data.

[0187] Optionally, when the eleventh submodule uses the first high-confidence pseudo-label data as a training sample increment sample of the support vector machine classifier, it is specifically used to:

[0188] For each category included in the first classification result, determine first high-confidence pseudo-label data corresponding to each category;

[0189] Perform inter-class data balancing processing on the first high-confidence pseudo-label data corresponding to each classification;

[0190] The first high-confidence pseudo-label data after inter-class balancing processing is used as a training sample increment sample of the support vector machine classifier.

[0191] The device for determining the type of forest disturbance provided in the embodiment of the present application combines small sample learning and the mutual guidance learning method of two base classifiers, random forest and support vector machine, to train the model. The user only needs to manually mark a small number of samples, and then the amount of sample data is expanded by iterative mutual guidance, which can save the time and labor cost of sample marking.

[0192] The embodiments of the present application provide Figure 4 The device for determining the type of forest disturbance shown enables Figure 1 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0193] An embodiment of the present invention further provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus.

[0194] Memory, used to store computer programs;

[0195] The processor is used to implement any one of the methods for determining the type of forest disturbance described in the above embodiments when executing the program stored in the memory.

[0196] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0197] The communication interface is used for communication between the above terminal and other devices.

[0198] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0199] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0200] In another embodiment provided by the present invention, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is executed on a computer, the computer implements the method for determining the forest disturbance type described in any one of the above embodiments.

[0201] In another embodiment of the present invention, a computer program product comprising instructions is provided. When the computer program product is executed on a computer, the computer implements the method for determining the forest disturbance type described in any one of the above embodiments.

[0202] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0203] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0204] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for determining the type of forest disturbance, characterized in that: The method comprises: Obtaining forest disturbance data from a land cover dataset; wherein the disturbance data includes: disturbance time and disturbance occurrence location; Performing feature analysis on the forest disturbance data to obtain forest disturbance data under each classification; Taking the forest disturbance data under each classification as basic data, the random forest classifier and the support vector machine classifier are iteratively amplified for sample data and trained for classifiers multiple times to obtain random forest classifiers and support vector machine classifiers that meet preset conditions; The forest disturbance data to be predicted is predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type.

2. The method according to claim 1, characterized in that The steps to obtain forest disturbance data from the land cover dataset include: The initial land cover dataset was subjected to forest masking to obtain forest-related data; The forest-related data is detected using a continuous change detection and classification algorithm CCDC to obtain a detection result, wherein the detection result includes: a disturbance caused by a forest changing into other land types and a disturbance caused by other land types changing into a forest; Based on the land cover dataset, disturbance pixels of the forest change to other land types are filtered to obtain forest disturbance data.

3. The method according to claim 1, characterized in that The step of performing feature analysis on the forest disturbance data to obtain forest disturbance data under each category includes: Analyzing the forest disturbance data based on the temporal characteristics, spatial characteristics, spectral characteristics and topographic characteristics of the subtropical evergreen forest disturbance classification to obtain the classification corresponding to each forest disturbance; The forest disturbance data of the same classification are divided into the same classification to obtain the forest disturbance data under each classification.

4. The method according to claim 1, characterized in that: The steps of using the forest disturbance data under each classification as basic data, iterating multiple times to amplify the sample data volume and train the classifier by mutual induction of the random forest classifier and the support vector machine classifier, and obtaining the random forest classifier and the support vector machine classifier that meet the preset conditions include: For each classification of forest disturbance data, a preset number of forest disturbance data are taken as sample data; Divide the sample data of each category into training samples and test samples according to a preset ratio; The training samples are used to train the random forest classifier and the support vector machine classifier respectively; Predicting the test samples corresponding to each classification based on the trained random forest classifier and support vector machine classifier, and determining the test accuracy of the random forest classifier and support vector machine classifier; When the test accuracy does not meet the preset conditions, classifying the unlabeled forest disturbance data according to the trained random forest classifier and support vector machine classifier to obtain a first classification result and a second classification result; Determining training sample increment samples of the random forest classifier and the support vector machine classifier according to the first classification result and the second classification result; The random forest classifier is trained using the incremental training samples corresponding to the random forest classifier, and the support vector machine classifier is trained using the incremental training samples corresponding to the support vector machine classifier, thereby completing a mutual training of the random forest classifier and the support vector machine classifier.

5. The method according to claim 4, characterized in that The step of determining the training sample increment samples of the random forest classifier and the support vector machine classifier according to the first classification result and the second classification result includes: Comparing the first classification result with the second classification result, and determining the classification result with a difference as pseudo-label data; Filtering out first high-confidence pseudo-label data from the pseudo-label data included in the first classification result, and using the first high-confidence pseudo-label data as a training sample increment sample of the support vector machine classifier; Second high-confidence pseudo-label data are screened out from the pseudo-label data included in the second classification result, and the second high-confidence pseudo-label data are used as training sample increment samples of the random forest classifier.

6. The method according to claim 5, characterized in that The step of selecting first high-confidence pseudo-label data from the pseudo-label data included in the first classification result includes: From the pseudo-label data included in the first classification result, first pseudo-label data having a classification probability value greater than a first preset probability threshold is selected; A confidence learning algorithm is used to filter out erroneous pseudo-label data in the first pseudo-label data to obtain first high-confidence pseudo-label data.

7. The method according to claim 5, characterized in that The step of using the first high-confidence pseudo-label data as a training sample increment sample of the support vector machine classifier comprises: For each category included in the first classification result, determine first high-confidence pseudo-label data corresponding to each category; Perform inter-class data balancing processing on the first high-confidence pseudo-label data corresponding to each classification; The first high-confidence pseudo-label data after inter-class balancing processing is used as a training sample increment sample of the support vector machine classifier.

8. A device for determining the type of forest disturbance, characterized in that The device comprises: An acquisition module is used to acquire forest disturbance data from a land cover data set; wherein the disturbance data includes: disturbance time and disturbance occurrence location; A classification module, used for performing feature analysis on the forest disturbance data to obtain forest disturbance data under each classification; A mutual guidance training module is used to use the forest disturbance data under each classification as basic data, and adopt a random forest classifier and a support vector machine classifier to iteratively amplify the sample data and train the classifier multiple times in a mutual guidance manner to obtain a random forest classifier and a support vector machine classifier that meet preset conditions; The prediction module is used to predict the forest disturbance data to be predicted based on the random forest classifier and the support vector machine classifier that meet the preset conditions to obtain the forest disturbance type.

9. The device according to claim 8, characterized in that The acquisition module comprises: The first submodule is used to perform forest masking on the initial land cover dataset to obtain forest-related data; The second submodule is used to detect the forest-related data using a continuous change detection and classification algorithm CCDC to obtain a detection result, wherein the detection result includes: a disturbance in which a forest changes into other land types and a disturbance in which other land types change into a forest; The third submodule is used to filter the disturbance pixels of the forest change into other land types based on the land cover dataset to obtain forest disturbance data.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the method for determining the forest disturbance type as described in any one of claims 1-7 when executing the program stored in the memory.

Images