Remote sensing sample annotation method based on geoscientific information and active learning

Abstract

The present invention discloses a remote sensing sample labeling method based on geoscience information and active learning, which belongs to the field of remote sensing image classification, and includes: obtaining a remote sensing sample set; performing geoscience calculation on the remote sensing sample set to obtain geoscience information; clustering the remote sensing sample set according to the geoscience information; obtaining a labeled sample set and an unlabeled sample set in combination with an active learning method; training a first classifier model through the labeled sample set, inputting the unlabeled sample set into the first classifier model for prediction, and screening in combination with geoscience information and sample query strategy to obtain a valuable sample set; after the valuable sample set is handed over to an expert for labeling, the valuable sample set labeled by the expert is added to the labeled sample set; labeling the unlabeled sample set through the first classifier model to obtain a labeling result. The labeling method of the present invention can improve the accuracy of labeling.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing image classification, and in particular to a remote sensing sample labeling method based on geoscientific information and active learning. Background Art

[0002] The present invention is aimed at the remote sensing sample annotation in large-area scenes. The traditional supervised learning method needs to label each sample, so it is difficult to be applied in the context of large-area scenes. Active learning is a method that can reduce the sample annotation cost while ensuring the accuracy of sample annotation. The traditional supervised learning method requires experts to label the samples. In fact, the expert's labeling process for training samples is usually completed based on the visual features of the scene. Therefore, if the samples are directly handed over to experts for labeling before screening, the consequence is that the experts will spend a lot of valuable time to fully label samples with similar information, which not only wastes a lot of manual resources, but also makes the information of the training set very redundant. This redundant information greatly reduces the training speed and even causes overfitting. Therefore, for satellite remote sensing images, we need a process of automatically defining an effective training set. The number of samples in this training set should be as small as possible and can effectively improve the accuracy of the classification model. Therefore, active learning comes into being. Active learning requires a very small number of labeled samples to perform initial training of the classifier. The number of these labeled samples is far less than the number required to fully train a classifier. Afterwards, a specific screening strategy is used to select a specific number of samples from the current samples to be labeled. These selected samples are manually labeled. Finally, these newly labeled samples are used for incremental training of the classifier.

[0003] However, even if active learning screening strategies are used to reduce the number of labeled samples at large regional or global scales, the number of samples that need to be manually labeled is still relatively large, resulting in very high labor costs, large amounts of data processing, and low accuracy of trained classifier models, making it difficult to complete sample labeling at large regional or global scales. The main reason is that existing active learning methods cannot fully utilize the information of remote sensing samples. Summary of the invention

[0004] The technical problem to be solved by the present invention is to comprehensively utilize the spatial characteristics and statistical characteristics of remote sensing samples, organically combine geological information with data mining methods, and increase the accuracy of sample annotation.

[0005] To achieve the above-mentioned purpose of the invention, the present invention provides a remote sensing sample annotation method based on geoscientific information and active learning, comprising:

[0006] S1 obtains a remote sensing sample set, which consists of multiple remote sensing samples. The remote sensing samples are divided into unlabeled samples and labeled samples;

[0007] S2 performs geoscientific calculations on the remote sensing sample set to obtain geoscientific information, wherein the geoscientific information includes elevation information, spectral information, texture information, shape information, and statistical measurement information;

[0008] S3 clusters the remote sensing sample set according to the geoscientific information to obtain k clusters and k cluster centers, where each cluster includes a cluster center, k ≥ 1;

[0009] S4 calculates the distance between each cluster center and the remote sensing samples in the corresponding cluster. Each cluster selects the remote sensing sample closest to the cluster center and the remote sensing sample farthest from the cluster center to obtain 2k remote sensing samples.

[0010] S5 gives the unlabeled samples in the 2k remote sensing samples to experts for labeling, combines the expert labeling results and the labeled samples in the remote sensing sample set into a labeled sample set, and divides the remote sensing sample set into a labeled sample set and an unlabeled sample set;

[0011] S6 performs model training on the first classifier model using the labeled sample set, and determines whether the conditions for terminating the training of the first classifier model are met:

[0012] If satisfied, the training ends and step S9 is executed;

[0013] If not satisfied, execute step S7;

[0014] S7 inputs the unlabeled sample set into the first classifier model for prediction, and screens it in combination with the geoscientific information and the sample query strategy to obtain a valuable sample set;

[0015] S8 After the valuable sample set is handed over to the expert for labeling, the valuable sample set labeled by the expert is added to the labeled sample set, and the unlabeled sample set is updated and then the process returns to step S6;

[0016] S9: annotate the unlabeled sample set using the first classifier model to obtain an annotation result.

[0017] In a specific embodiment of the present invention, step S3 includes:

[0018] S31 obtains the location information of each remote sensing sample and constructs a distance calculation strategy based on the geoscientific information. The distance calculation strategy includes a spatial distance method and a feature distance method.

[0019] S32 obtains k initial cluster centers based on the distance calculation strategy;

[0020] S33 combines the location information of remote sensing samples and the distance calculation strategy to iteratively optimize the k initial cluster centers to obtain k clusters and k cluster centers.

[0021] In a specific embodiment of the present invention, step S32 includes:

[0022] S321 randomly selects a remote sensing sample from the remote sensing sample set, uses the remote sensing sample as an initial cluster center, and adds it to the initial cluster center set;

[0023] S322 calculates the distances between a single remote sensing sample and all the initial cluster centers based on the distance calculation strategy, takes the largest distance as the first distance of the remote sensing sample, sorts the first distances of all remote sensing samples from large to small, selects the remote sensing sample with the largest first distance as the new initial cluster center, and adds it to the initial cluster center set;

[0024] S323 Repeat step S322 until the number of initial cluster centers in the initial cluster center set reaches k.

[0025] In a specific embodiment of the present invention, step S33 includes:

[0026] S331 obtains the coordinate value of the remote sensing sample according to the position information of the remote sensing sample;

[0027] S332 calculates the distances between a single remote sensing sample and the k initial cluster centers based on the distance calculation strategy, and uses the smallest distance as the second distance of the remote sensing sample;

[0028] S333: forming an initial cluster cluster with a single initial cluster center and remote sensing samples whose distance from the initial cluster center is the second distance thereof, and using the initial cluster center as the initial cluster center of the cluster cluster, thereby obtaining initial k cluster clusters and initial k cluster centers;

[0029] S334 calculates the average value of the coordinate values ​​of all remote sensing samples in the current single cluster, and calculates the difference between the coordinate value of each remote sensing sample and the average value, and uses the remote sensing sample corresponding to the coordinate value with the smallest difference as a new cluster center to obtain new k cluster centers;

[0030] S335 forms a new cluster with a single new cluster center and remote sensing samples whose distance from the cluster center is the second distance thereof, and obtains k new cluster clusters;

[0031] S336 calculates the distance between each remote sensing sample and the corresponding new cluster center according to the distance calculation strategy, and calculates the sum of squares of all distances to obtain the sum of squares of errors of the new k clusters;

[0032] S337 Iterate and execute steps S334-S336. Each iteration obtains k clusters and their k cluster centers, and the sum of squared errors of the k clusters. The change value is calculated based on the sum of squared errors of two adjacent iterations to determine whether the change value meets the iteration stopping condition. If so, the iteration is stopped to obtain the final k clusters and k cluster centers.

[0033] In a specific embodiment of the present invention, the distance calculation strategy is:

[0034] Select two remote sensing samples to be calculated as the first sample and the second sample;

[0035] The spatial distance between the first sample and the second sample is obtained according to the spatial distance method ;

[0036] The characteristic distance between the first sample and the second sample is obtained according to the characteristic distance method ;

[0037] Will and Perform normalization processing to obtain the normalized processing result and ,in and The range of is [0,1];

[0038] calculate and The sum of is taken as the distance between the first sample and the second sample.

[0039] In a specific embodiment of the present invention, the spatial distance method is:

[0040] Construct a Delaunay triangulation network {Del} according to the location information of the remote sensing samples, {Del} includes multiple Delaunay triangles, and each Delaunay triangle includes three vertices and adjacent edges;

[0041] Get the Delaunay triangle Del of the first sample and the second sample in the Delaunay triangulation {Del} 1 and Del 2 ;

[0042] Get Del 1 Get Del from the vertex set {Node1} on its adjacent edge 2 The set of vertices on its adjacent edges {Node2};

[0043] According to the coordinates of each vertex in {Node1} and {Node2}, get the two vertices Node with the farthest spatial position 1 and Node2 ;

[0044] Calculate Node based on spatial topological relationship 1 and Node 2 The distance between the first sample and the second sample .

[0045] In a specific embodiment of the present invention, the adjacent sides of a Delaunay triangle are the sides shared by the Delaunay triangle and other Delaunay triangles, and the number of adjacent sides of each Delaunay triangle is different.

[0046] In a specific embodiment of the present invention, the characteristic distance method is:

[0047] According to the geoscientific information, the geoscientific information vector f of the first sample and the second sample is obtained 1 and f 2 ;

[0048] Calculate f 1 and f 2 The Euclidean distance of the first sample and the second sample is :

[0049] .

[0050] In a specific embodiment of the present invention, step S7 includes:

[0051] S71 calculates the information entropy and probability density of each unlabeled sample in the unlabeled sample set, and calculates the product of the information entropy and probability density of each unlabeled sample, and selects the unlabeled samples by combining the product and the difference constraint condition to obtain key samples;

[0052] S72 obtains labeled samples in the same cluster as the key sample as important samples;

[0053] S73 calculates the characteristic distance between each key sample and its corresponding important sample as the third distance, and adds the key samples whose third distance is greater than the distance threshold to the value sample set.

[0054] In a specific embodiment of the present invention, it is characterized in that:

[0055] The elevation information includes DEM information, ground slope information, and terrain roughness information;

[0056] The spectral information includes normalized vegetation index and enhanced vegetation index;

[0057] The texture information includes grayscale co-occurrence matrix information, grayscale run length matrix information, and neighborhood grayscale difference matrix information;

[0058] The shape information includes rectangularity, elongation, major axis length, and longest diameter;

[0059] The statistical measurement information includes maximum value, minimum value, range, and skewness.

[0060] The present invention provides a remote sensing sample annotation method based on geoscientific information and active learning. In summary, due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0061] (1) The present invention clusters samples based on geological information, and can comprehensively utilize the spatial and statistical characteristics of remote sensing samples to obtain clusters with continuous characteristics and space, and perform initial sample selection and labeling from the clusters. Compared with the existing active learning methods, the present invention can better ensure the polymorphism of samples.

[0062] (2) The present invention can reduce the cost of sample annotation and quickly improve the classification effect of the classifier model.

[0063] (3) The present invention utilizes a sample query strategy combined with geological information to screen unlabeled samples and obtain a valuable sample set, which can obtain valuable samples that are both representative and informative. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0065] Figure 1 It is a method flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0067] like Figure 1 As shown, this embodiment provides a remote sensing sample labeling method based on geoscientific information and active learning, including:

[0068] S1 obtains a remote sensing sample set, which consists of multiple remote sensing samples. The remote sensing samples are divided into unlabeled samples and labeled samples.

[0069] Firstly, multiple remote sensing samples are obtained, including unlabeled samples and labeled samples, to form a remote sensing sample set, where the number of unlabeled samples is much larger than the number of labeled samples.

[0070] S2 performs geoscientific calculations on the remote sensing sample set to obtain geoscientific information, wherein the geoscientific information includes elevation information, spectral information, texture information, shape information, and statistical measurement information.

[0071] Among them, the elevation information includes DEM information, ground slope information, and terrain roughness information; the spectral information includes normalized vegetation index and enhanced vegetation index; the texture information includes grayscale co-occurrence matrix information, grayscale run length matrix information, and neighborhood grayscale difference matrix information; the shape information includes rectangularity, elongation, major axis length, and longest diameter; the statistical measurement information includes maximum value, minimum value, range, and skewness.

[0072] Specifically, geological information is used to reflect the spatial location distribution characteristics of ground objects in remote sensing samples, the attributes of ground objects, and other geographical information. The geological information of remote sensing samples can be obtained through geological calculation methods, such as geological data extraction and analysis methods.

[0073] S3 clusters the remote sensing sample set according to the geoscientific information to obtain k clusters and k cluster centers, where each cluster includes a cluster center, k≥1.

[0074] Specifically, in one embodiment of the present invention, step S3 includes:

[0075] S31 obtains the location information of each remote sensing sample and constructs a distance calculation strategy based on the geological information. The distance calculation strategy includes a spatial distance method and a feature distance method.

[0076] In one embodiment of the present invention, the distance calculation strategy is:

[0077] Two remote sensing samples to be calculated are selected as the first sample and the second sample.

[0078] The spatial distance between the first sample and the second sample is obtained according to the spatial distance method .

[0079] Specifically, the spatial distance method is:

[0080] A Delaunay triangulation {Del} is constructed according to the location information of the remote sensing samples. {Del} includes multiple Delaunay triangles, and each Delaunay triangle includes three vertices and adjacent edges.

[0081] It should be noted that the Delaunay triangulation is a collection of a series of connected but non-overlapping Delaunay triangles, and the circumscribed circles of these Delaunay triangles do not contain any other points in the area. When constructing a Delaunay triangulation based on the location information of remote sensing samples, the geographical location of the remote sensing samples at the time of imaging is used, such as spatial coordinates, longitude and latitude, etc. In the Delaunay triangulation, each remote sensing sample falls inside the corresponding Delaunay triangle.

[0082] Each Delaunay triangle has three vertices and three edges. When a Delaunay triangle is connected to another Delaunay triangle, the two Delaunay triangles will share the same edge. The edge shared by the Delaunay triangle and the other Delaunay triangle is used as the adjacent edge of the Delaunay triangle. There are many situations for a Delaunay triangle. When it is connected to another Delaunay triangle, it has one adjacent edge. When it is connected to two other Delaunay triangles, it has two adjacent edges. When it is connected to three other Delaunay triangles, it has three adjacent edges. Therefore, the number of adjacent edges of each Delaunay triangle is different.

[0083] Get the Delaunay triangle Del of the first sample and the second sample in the Delaunay triangulation {Del} 1 and Del 2 .

[0084] Get Del 1 Get Del from the vertex set {Node1} on its adjacent edge 2 The set of vertices on its adjacent edges {Node2}.

[0085] According to the coordinates of each vertex in {Node1} and {Node2}, get the two vertices Node with the farthest spatial position 1 and Node 2 .

[0086] Specifically, the spatial position between every two vertices is obtained according to the position of the coordinates of each vertex in the spatial coordinate system.

[0087] Calculate Node based on spatial topological relationship 1 and Node 2 The distance between the first sample and the second sample .

[0088] Specifically, Node 1 and Node2 The distance between them is a spatial distance, which cannot be calculated based on the two-dimensional plane method. Therefore, this embodiment adopts the spatial topology calculation method and uses the adjacent edges of the Delaunay triangle to obtain the distance between the two points. For example, Node 1 Del 1 With Node 2 Del 2 There are two Delaunay triangles between them, denoted as Del 3 and Del 4 , Del 1 With Del 3 Connected, Del 3 With Del 4 Connected, Del 4 With Del 2 Connected, from Node 1 Start out and follow Del 1 The adjacent edge of Del 3 The adjacent edge of Del 4 The adjacent edge of Del 2 The adjacent edge to Node 1 So far, the shortest spatial path between two points is obtained, and the distance between the two points is obtained through topological calculation.

[0089] The characteristic distance between the first sample and the second sample is obtained according to the characteristic distance method .

[0090] Specifically, the feature distance method is:

[0091] According to the geoscientific information, the geoscientific information vector f of the first sample and the second sample is obtained 1 and f 2 ;

[0092] Calculate f 1 and f 2 The Euclidean distance of the first sample and the second sample is :

[0093]

[0094] Among them, the geoscientific information vector is extracted and calculated based on the geoscientific information, and specifically can be one or more of the elevation information vector, the spectral information vector, the texture information vector, the shape information vector, and the statistical measurement information vector. When there are multiple types, the multiple vectors can be spliced ​​or fused to obtain the geoscientific information vector.

[0095] Will and Perform normalization processing to obtain the normalized processing result and ,in and The range of is [0,1].

[0096] calculate and The sum of is taken as the distance between the first sample and the second sample.

[0097] S32 obtains k initial cluster centers based on the distance calculation strategy.

[0098] Specifically, step S32 may include:

[0099] S321 randomly selects a remote sensing sample from the remote sensing sample set, uses the remote sensing sample as an initial cluster center, and adds it to the initial cluster center set.

[0100] S322 calculates the distances between a single remote sensing sample and all the initial cluster centers based on the distance calculation strategy, takes the largest distance as the first distance of the remote sensing sample, sorts the first distances of all remote sensing samples from large to small, selects the remote sensing sample with the largest first distance as the new initial cluster center, and adds it to the initial cluster center set.

[0101] S323 Repeat step S322 until the number of initial cluster centers in the initial cluster center set reaches k.

[0102] Specifically, step S32 is described with an embodiment:

[0103] The remote sensing sample set is denoted as X={X 1 ,X 2 ,...,X n}, n is the number of remote sensing samples in the remote sensing sample set, and a remote sensing sample X is randomly selected from X i , respectively calculate the remaining n-1 remote sensing samples {X 1 ,X 2 ,...,X i-1 ,X i+1 ,...,X n} and X i The distance between them is taken as {X 1 ,X 2 ,...,X i-1 ,X i+1 ,...,X n} their respective first distances, for {X 1 ,X 2 ,...,X i-1 ,X i+1 ,...,Xn} are sorted from large to small, and the first remote sensing sample is selected. Assume that the remote sensing sample is X 1 , then X 1 and X i All of them are used as initial cluster centers, and an initial cluster center set is constructed.

[0104] Calculate the remaining n-2 remote sensing samples {X 2 ,...,X i-1 ,X i+1 ,...,X n} respectively with X i and X 1 The largest distance is taken as the first distance of the corresponding remote sensing sample, for example, X 2 With X i The distance ratio X 2 With X 1 The distance between 2 The first distance is its distance from X i distance, and similarly {X 2 ,...,X i-1 ,X i+1 ,...,X n} are sorted from large to small, and the first remote sensing sample is selected as the new initial cluster center and added to the initial cluster center set.

[0105] The initial cluster centers are selected in sequence according to the rule described above until the number of initial cluster centers in the initial cluster center set reaches k. In this embodiment, k can be 6.

[0106] S33 combines the location information of remote sensing samples and the distance calculation strategy to iteratively optimize the k initial cluster centers to obtain k clusters and k cluster centers.

[0107] In one embodiment of the present invention, step S33 includes:

[0108] S331 obtains the coordinate value of the remote sensing sample according to the position information of the remote sensing sample.

[0109] Specifically, the location information of the remote sensing sample can be obtained based on the metadata of the remote sensing sample, which is the data obtained when the remote sensing sample is imaged, and refers to the actual geographical location information of the remote sensing sample when it is imaged. The coordinate value of the remote sensing sample in the global geographic coordinate system can be obtained based on the location information.

[0110] S332 calculates the distances between a single remote sensing sample and the k initial cluster centers based on the distance calculation strategy, and takes the smallest distance as the second distance of the remote sensing sample.

[0111] Specifically, the distance between each remote sensing sample and k initial clustering centers is calculated, that is, each remote sensing sample can obtain k distances, and the smallest of these k distances is used as the second distance of the corresponding remote sensing sample.

[0112] S333 forms an initial cluster cluster with a single initial cluster center and remote sensing samples whose distance to the initial cluster center is the second distance thereof, and uses the initial cluster center as the initial cluster center of the cluster cluster, thereby obtaining initial k cluster clusters and initial k cluster centers.

[0113] Specifically, in an initial clustering cluster, there is an initial clustering center and multiple remote sensing samples. In the initial clustering cluster, the distance between each remote sensing sample and the initial clustering center is its second distance. The initial clustering center is recorded as the initial clustering center of the initial clustering cluster. Finally, the initial k clusters and the initial k clustering centers are obtained.

[0114] S334 averages the coordinate values ​​of all remote sensing samples within the current single cluster, and calculates the difference between the coordinate value of each remote sensing sample and the average value, and uses the remote sensing sample corresponding to the coordinate value with the smallest difference as the new cluster center to obtain new k cluster centers.

[0115] Specifically, taking the current cluster as the target, the average coordinate values ​​of all remote sensing samples in a single cluster are calculated. It should be noted that all remote sensing samples mentioned here refer to remote sensing samples other than the current cluster center. Then the difference between the coordinate value of each remote sensing sample and the average value is calculated, and the remote sensing sample with the smallest difference is used as the new cluster center, that is, the cluster center is replaced. According to the above steps, all current cluster centers are replaced to obtain new k cluster centers.

[0116] S335 forms a new cluster with a single new cluster center and remote sensing samples whose distance from the cluster center is the second distance, and obtains k new clusters.

[0117] Specifically, after obtaining the new k cluster centers, k new clusters are still formed around the new k cluster centers according to the second distance to complete the update of the clusters.

[0118] S336 calculates the distance between each remote sensing sample and the corresponding new cluster center according to the distance calculation strategy, and calculates the sum of squares of all distances to obtain the sum of squares of errors of the new k clusters.

[0119] It can be understood that, taking a single new cluster as the object, the distance between the remote sensing sample and the corresponding new cluster center is calculated, that is, the second distance of the remote sensing sample, and the square sum of the second distances of the remote sensing samples of all new clusters is calculated together to obtain the square sum of the errors of the new k clusters, that is, the square sum of the errors of the new k clusters is a value, and its calculation formula is as follows:

[0120]

[0121] Among them, SSE represents the sum of squared errors, k is the number of clusters, and m is the i is the number of remote sensing samples in the i-th cluster, is the distance between the remote sensing sample in the i-th cluster and the cluster center.

[0122] S337 Iterate and execute steps S334-S336. Each iteration obtains k clusters and their k cluster centers, and the sum of squared errors of the k clusters. The change value is calculated based on the sum of squared errors of two adjacent iterations to determine whether the change value meets the iteration stopping condition. If so, the iteration is stopped to obtain the final k clusters and k cluster centers.

[0123] Specifically, the iteration stopping condition may be that the change value between the sum of squared errors obtained from two adjacent iterations is 0, that is, the sum of squared errors has been minimized. Alternatively, the iteration stopping condition may be that the maximum number of iterations has been reached, for example, if the maximum number of iterations is 6, then the iteration is stopped after 6 iterations. Alternatively, the iteration stopping condition may be that the change value reaches a threshold value, which may be set to 0.2.

[0124] S4 calculates the distance between each cluster center and the remote sensing samples in the corresponding cluster. Each cluster selects the remote sensing sample closest to the cluster center and the remote sensing sample farthest from the cluster center to obtain 2k remote sensing samples.

[0125] Specifically, taking a single cluster as the object, the distance between each remote sensing sample in the cluster and the cluster center is calculated. The distance is still calculated according to the distance calculation strategy. The distances are sorted from large to small, and the first remote sensing sample and the last remote sensing sample are selected. Finally, 2k remote sensing samples can be selected from the k clusters.

[0126] S5 gives the unlabeled samples in the 2k remote sensing samples to experts for labeling, combines the expert labeling results and the labeled samples in the remote sensing sample set into a labeled sample set, and divides the remote sensing sample set into a labeled sample set and an unlabeled sample set.

[0127] Specifically, if the selected 2k remote sensing samples include unlabeled samples, they will first be handed over to experts for labeling and converted into labeled samples. Then all remote sensing samples will be re-divided according to whether they are labeled or not, obtaining labeled sample sets and unlabeled sample sets.

[0128] S6 performs model training on the first classifier model using the labeled sample set, and determines whether the conditions for terminating the training of the first classifier model are met:

[0129] If satisfied, the training ends and step S9 is executed;

[0130] If not satisfied, execute step S7.

[0131] S7 inputs the unlabeled sample set into the first classifier model for prediction, and screens it in combination with the geological information and sample query strategy to obtain a valuable sample set.

[0132] Specifically, step S7 includes:

[0133] S71 calculates the information entropy and probability density of each unlabeled sample in the unlabeled sample set, and calculates the product of the information entropy and probability density of each unlabeled sample, and combines the product with the difference constraint condition to screen the unlabeled samples to obtain key samples.

[0134] S72 obtains labeled samples in the same cluster as the key sample as important samples.

[0135] S73 calculates the characteristic distance between each key sample and its corresponding important sample as the third distance, and adds the key samples whose third distance is greater than the distance threshold to the value sample set.

[0136] Specifically, step S7 uses active learning to query samples. This embodiment uses information entropy to measure the informativeness of unlabeled samples, which is defined as follows:

[0137]

[0138] in, It represents the probability that the unlabeled sample x belongs to the jth category.

[0139] In addition, this embodiment chooses to use probability density to estimate the representativeness of unlabeled samples, which is defined as follows:

[0140]

[0141] Among them, m is the number of unlabeled samples, is the Gaussian kernel function.

[0142] Calculate the product of information entropy and probability density of each unlabeled sample, and sort them from small to large. Select the first unlabeled sample directly as the key sample, and the remaining unlabeled samples need to meet the difference constraint. The difference constraint refers to the difference between the unlabeled sample currently queried and the existing key samples. The specific difference can be measured according to the difference between the maximum information entropy and the product of probability density, that is, the maximum value of the difference between the product of the unlabeled sample currently queried and the product of each existing key sample is taken as the difference of the unlabeled sample. The difference must be lower than the difference threshold, which can be set to 0.1.

[0143] After the key samples are obtained through query, the corresponding labeled samples are obtained according to the cluster where each key sample is located, and these labeled samples are used as important samples corresponding to the key samples.

[0144] According to the geoscientific information, the geoscientific information vectors of key samples and important samples are obtained, and then the characteristic distance between a single key sample and its corresponding important sample is calculated based on the characteristic distance method, and the largest characteristic distance is selected as the third distance of the key sample. The third distances of all key samples are compared with the distance threshold, and the key samples greater than the distance threshold are added to the valuable sample set. Among them, the distance threshold can be set to 0.5.

[0145] S8 After the valuable sample set is handed over to the expert for labeling, the valuable sample set labeled by the expert is added to the labeled sample set, and the unlabeled sample set is updated and then the process returns to step S6;

[0146] S9: annotate the unlabeled sample set using the first classifier model to obtain an annotation result.

[0147] The above implementation modes are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention. The patent protection scope of the present invention should be defined by the claims.

Claims

1. A remote sensing sample annotation method based on geoscientific information and active learning, characterized in that: The following steps are involved: S1 obtains a remote sensing sample set, which consists of multiple remote sensing samples. The remote sensing samples are divided into unlabeled samples and labeled samples; S2 performs geoscientific calculations on the remote sensing sample set to obtain geoscientific information, wherein the geoscientific information includes elevation information, spectral information, texture information, shape information, and statistical measurement information; S3 clusters the remote sensing sample set according to the geoscientific information to obtain k clusters and k cluster centers, where each cluster includes a cluster center, k ≥ 1; S4 calculates the distance between each cluster center and the remote sensing samples in the corresponding cluster. Each cluster selects the remote sensing sample closest to the cluster center and the remote sensing sample farthest from the cluster center to obtain 2k remote sensing samples. S5 gives the unlabeled samples in the 2k remote sensing samples to experts for labeling, combines the expert labeling results and the labeled samples in the remote sensing sample set into a labeled sample set, and divides the remote sensing sample set into a labeled sample set and an unlabeled sample set; S6 performs model training on the first classifier model using the labeled sample set, and determines whether the conditions for terminating the training of the first classifier model are met: If satisfied, the training ends and step S9 is executed; If not satisfied, execute step S7; S7 inputs the unlabeled sample set into the first classifier model for prediction, and screens it in combination with the geoscientific information and the sample query strategy to obtain a valuable sample set; S8 After the valuable sample set is handed over to the expert for labeling, the valuable sample set labeled by the expert is added to the labeled sample set, and the unlabeled sample set is updated and then the process returns to step S6; S9 labels the unlabeled sample set by using the first classifier model to obtain a labeling result; Wherein, the step S3 comprises: S31 obtains the location information of each remote sensing sample and constructs a distance calculation strategy based on the geoscientific information. The distance calculation strategy includes a spatial distance method and a feature distance method. S32 obtains k initial cluster centers based on the distance calculation strategy; S33 combines the location information of remote sensing samples and the distance calculation strategy to iteratively optimize the k initial cluster centers to obtain k clusters and k cluster centers.

2. A remote sensing sample labeling method based on geoscientific information and active learning as claimed in claim 1, characterized in that: Step S32 includes: S321 randomly selects a remote sensing sample from the remote sensing sample set, uses the remote sensing sample as an initial cluster center, and adds it to the initial cluster center set; S322 calculates the distances between a single remote sensing sample and all the initial cluster centers based on the distance calculation strategy, takes the largest distance as the first distance of the remote sensing sample, sorts the first distances of all remote sensing samples from large to small, selects the remote sensing sample with the largest first distance as the new initial cluster center, and adds it to the initial cluster center set; S323 repeats step S322 until the number of initial cluster centers in the initial cluster center set reaches k.

3. A remote sensing sample labeling method based on geoscientific information and active learning as claimed in claim 2, characterized in that: Step S33 includes: S331 obtains the coordinate value of the remote sensing sample according to the position information of the remote sensing sample; S332 calculates the distances between a single remote sensing sample and the k initial cluster centers based on the distance calculation strategy, and uses the smallest distance as the second distance of the remote sensing sample; S333: forming an initial cluster cluster with a single initial cluster center and remote sensing samples whose distance from the initial cluster center is the second distance thereof, and using the initial cluster center as the initial cluster center of the cluster cluster, thereby obtaining initial k cluster clusters and initial k cluster centers; S334 calculates the average value of the coordinate values ​​of all remote sensing samples in the current single cluster, and calculates the difference between the coordinate value of each remote sensing sample and the average value, and uses the remote sensing sample corresponding to the coordinate value with the smallest difference as a new cluster center to obtain new k cluster centers; S335 forms a new cluster with a single new cluster center and remote sensing samples whose distance from the cluster center is the second distance thereof, and obtains k new cluster clusters; S336 calculates the distance between each remote sensing sample and the corresponding new cluster center according to the distance calculation strategy, and calculates the sum of squares of all distances to obtain the sum of squares of errors of the new k clusters; S337 iterates and executes steps S334-S336. Each iteration obtains k clusters and their k cluster centers, and the sum of squared errors of the k clusters. The change value is calculated based on the sum of squared errors of two adjacent iterations to determine whether the change value meets the iteration stopping condition. If so, the iteration is stopped to obtain the final k clusters and k cluster centers.

4. A remote sensing sample annotation method based on geoscientific information and active learning as claimed in claim 1, characterized in that: The distance calculation strategy is: Select two remote sensing samples to be calculated as the first sample and the second sample; The spatial distance d between the first sample and the second sample is obtained according to the spatial distance method s ; According to the characteristic distance method, the characteristic distance d between the first sample and the second sample is obtained Eu ; D s and d Eu Perform normalization processing to obtain the normalized processing result d' s and d' Eu , where d' s and d' Eu The range of is [0,1]; Calculate d' s and d' Eu The sum of is taken as the distance between the first sample and the second sample.

5. A remote sensing sample annotation method based on geoscientific information and active learning as claimed in claim 4, characterized in that: The spatial distance method is: Construct a Delaunay triangulation network {Del} according to the location information of the remote sensing samples, {Del} includes multiple Delaunay triangles, and each Delaunay triangle includes three vertices and adjacent edges; Get the Delaunay triangle Del of the first sample and the second sample in the Delaunay triangulation {Del} 1 and Del 2 ; Get Del 1 Get Del from the vertex set {Node1} on its adjacent edge 2 The set of vertices on its adjacent edges {Node2}; According to the coordinates of each vertex in {Node1} and {Node2}, the two vertices Node1 and Node2 with the farthest spatial positions are obtained; Calculate the distance between Node1 and Node2 according to the spatial topological relationship, and use it as the spatial distance d between the first sample and the second sample. s .

6. A remote sensing sample annotation method based on geoscientific information and active learning as claimed in claim 5, characterized in that: The adjacent edges of a Delaunay triangle are the edges shared by the Delaunay triangle and other Delaunay triangles, and the number of adjacent edges of each Delaunay triangle is different.

7. A remote sensing sample annotation method based on geoscientific information and active learning as claimed in claim 4, characterized in that: The feature distance method is: According to the geoscientific information, the geoscientific information vector f of the first sample and the second sample is obtained 1 and f 2 ; Calculate f 1 and f 2 The Euclidean distance of the first sample and the second sample is taken as the characteristic distance d Eu :

8. A remote sensing sample annotation method based on geoscientific information and active learning as claimed in claim 7, characterized in that: Step S7 includes: S71 calculates the information entropy and probability density of each unlabeled sample in the unlabeled sample set, and calculates the product of the information entropy and probability density of each unlabeled sample, and selects the unlabeled samples by combining the product and the difference constraint condition to obtain key samples; S72 obtains labeled samples in the same cluster as the key sample as important samples; S73 calculates the characteristic distance between each key sample and its corresponding important sample as the third distance, and adds the key samples whose third distance is greater than the distance threshold to the value sample set.

9. The remote sensing sample annotation method based on geoscientific information and active learning as claimed in claim 1, characterized in that: The elevation information includes DEM information, ground slope information, and terrain roughness information; The spectral information includes normalized vegetation index and enhanced vegetation index; The texture information includes grayscale co-occurrence matrix information, grayscale run length matrix information, and neighborhood grayscale difference matrix information; The shape information includes rectangularity, elongation, major axis length, and longest diameter; The statistical measurement information includes maximum value, minimum value, range, and skewness.

Images