Geographic region type determination method, electronic equipment and storage medium

Abstract

The invention relates to the technical field of image processing, in particular to a geographic region type determination method, electronic equipment and a storage medium, and the method comprises the steps: dividing each remote sensing image of a target geographic region into a plurality of sub-geographic region images, and for the plurality of sub-geographic region images corresponding to each sub-geographic region, determining the types of the sub-geographic region images; performing plot attribute feature extraction on each sub-geographic region image to obtain a target feature vector corresponding to each sub-geographic region, performing clustering processing on a feature vector sample corresponding to each geographic region sample, and training a plurality of preset classification models by taking each obtained geographic region sample cluster as a sub-sample set, determining a target classification model corresponding to each geographic region sample cluster, finding a target sample cluster corresponding to the sub-geographic region, and finally determining a geographic region type of the sub-geographic region; the classification model with the highest accuracy for the geographic region features can be screened out, and then the classification accuracy of the geographic region types is improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method for determining a geographical area type, an electronic device and a storage medium. Background Art

[0002] With the increasing emphasis on food security, the monitoring of non-agricultural and non-grain land is increasingly valued. Carrying out non-agricultural and non-grain land monitoring is crucial to the effective issuance of farmland subsidies, and can also promote and ensure the sustainable development of farmland. The subsidized land is mainly grain fields, which refer to relatively concentrated and contiguous plots based on permanent basic farmland, with good water and soil resources, relatively complete infrastructure, and crops such as wheat, corn, and rice.

[0003] At present, remote sensing technology is usually used to monitor cultivated land areas. High-resolution satellites are launched into space to obtain qualified images, and the information of cultivated land plots is determined by remote sensing means. However, when processing image data, due to the different characteristics of different land types, the judgment accuracy of plots of different land types varies. Therefore, how to improve the judgment accuracy and realize the accurate division of plot types is very meaningful for the issuance of cultivated land subsidies. Summary of the invention

[0004] In response to the above technical problems, the present invention provides a method for determining geographic area types, an electronic device and a storage medium, which can screen out a classification model with the highest accuracy for geographic area characteristics, thereby improving the classification accuracy of geographic area types.

[0005] According to a first aspect of the present invention, there is provided a method for determining a geographical area type, comprising the following steps: Based on several remote sensing images of the target geographical area acquired within a preset time period, a multi-scale segmentation model is used to divide each remote sensing image of the target geographical area into several sub-geographical area images to obtain several sub-geographical area images corresponding to each sub-geographical area.

[0006] For a plurality of sub-geographical area images corresponding to each sub-geographical area, plot attribute features are extracted from each sub-geographical area image to obtain a target feature vector corresponding to each sub-geographical area.

[0007] Based on several geographic area samples obtained within a preset time period, the feature vector samples corresponding to each geographic area sample are clustered, and each obtained geographic area sample cluster is used as a sub-sample set to train several preset classification models, and the target classification model corresponding to each geographic area sample cluster is determined according to the geographic area type output result of each preset classification model.

[0008] According to the target feature vector corresponding to each sub-geographical area, a target sample cluster corresponding to each sub-geographical area is determined from a plurality of geographic area sample clusters.

[0009] The target feature vector corresponding to any sub-geographical area is input into the target classification model corresponding to the target sample cluster, and the geographical area type of any sub-geographical area is determined according to the output result of the target classification model.

[0010] According to a second aspect of the present invention, a non-transitory computer-readable storage medium is provided, in which at least one instruction or at least one program is stored, and the at least one instruction or the at least one program is loaded and executed by a processor to implement the above-mentioned method for determining the geographic area type.

[0011] According to a third aspect of the present invention, there is provided an electronic device comprising a processor and the above-mentioned non-transitory computer-readable storage medium.

[0012] The present invention has at least the following beneficial effects: The present invention provides a method for determining the type of a geographic region. First, each remote sensing image of a target geographic region is divided into a plurality of sub-geographic region images to obtain a plurality of sub-geographic region images corresponding to each sub-geographic region. Then, plot attribute features are extracted from each sub-geographic region image to obtain a target feature vector corresponding to the sub-geographic region. The target feature vector includes a plurality of feature dimensions that can be used to distinguish different geographic region types, so that the identified type of land objects is more comprehensive and accurate. Then, feature vector samples corresponding to each geographic region sample are clustered, and each acquired geographic region sample cluster is used as a sub-sample set to train a plurality of preset classification models to determine the target classification model corresponding to each geographic region sample cluster. In this process, the gap between the test result of each preset classification model and the actual geographic region type is considered to make the screened target classification model more suitable for the corresponding geographic region sample cluster. According to the geographic region sample cluster that is most similar to the sub-geographic region, the geographic region type of the sub-geographic region is finally determined. Therefore, the present invention can screen out the classification model with the highest accuracy for the geographic region features, thereby improving the classification accuracy of the geographic region type. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 A flowchart of a method for determining a geographic area type provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0016] The embodiment of the present invention provides a method for determining the type of a geographical area. Figure 1 As shown, the method comprises the following steps: S100, based on a plurality of remote sensing images of the target geographical area acquired within a preset time period, a multi-scale segmentation model is used to divide each remote sensing image of the target geographical area into a plurality of sub-geographical area images, so as to acquire a plurality of sub-geographical area images corresponding to each sub-geographical area.

[0017] Preferably, the preset time period is a time period during which several pre-selected crops are in the growth stage. For example, wheat is planted in November and harvested in May-June, and its intermediate growth stage can be used as the preset time period. Rice is planted in summer and harvested in autumn, and the intermediate growth stage of rice can be used as another preset time period. It should be noted that according to the subsidy issuance principle, subsidies can be issued as long as there is farming within the year. Therefore, images can be taken during the growing seasons of different crops, and the plots that need subsidies can be screened out based on the shooting results of each preset time period.

[0018] Furthermore, the edge detection algorithm in the multi-scale segmentation model is used to divide each remote sensing image into several sub-geographical area images; it can be understood that: according to the different types of crops or landforms in the target geographical area, the target geographical area is divided into several plots.

[0019] Specifically, a plurality of sub-geographical area images corresponding to each sub-geographical area are obtained through the following steps: S101, based on the number of sub-geographical area images corresponding to each remote sensing image, the number with the highest frequency among the several numbers is used as the target number corresponding to the sub-geographical area images, so as to screen out several remote sensing images corresponding to the target number. For example, since several remote sensing images of the target geographical area are obtained, some of them are not clear, resulting in different numbers and areas of the segmented sub-geographical area images, and therefore screening is required.

[0020] S102, for the screened remote sensing images, obtain the image center point coordinates of each sub-geographical area image in each remote sensing image, and cluster all the image center point coordinates according to the target number to obtain the target number of image center point clusters; it can be understood as: clustering several sub-geographical area images belonging to the same sub-geographical area into one class.

[0021] Furthermore, the k-means clustering model is used to cluster the coordinates of all image center points, where the value of k is the value of the target number.

[0022] S103, for any image center point cluster, delete the sub-geographic area images corresponding to the image center point coordinates whose distance from the cluster center of the image center point cluster is greater than a first preset distance threshold, and use the sub-geographic area images corresponding to the remaining several image center point coordinates in the image center point cluster as several sub-geographic area images of the same sub-geographic area.

[0023] As mentioned above, considering the problem of remote sensing image clarity, the remote sensing images were screened twice. First, the remote sensing images were screened for the first time according to the number of divided sub-geographical areas, and the remote sensing images obtained were clearer. Then, according to the center points of several sub-geographical area images of each sub-geographical area, the sub-geographical area images with large center point deviations were deleted, so that more accurately divided sub-geographical area images could be obtained, thereby making the extracted features of the sub-geographical areas closer to the actual situation.

[0024] S200 , for a plurality of sub-geographical area images corresponding to each sub-geographical area, extract land parcel attribute features from each sub-geographical area image to obtain a target feature vector corresponding to each sub-geographical area.

[0025] Preferably, the characteristic dimensions corresponding to the target characteristic vector include normalized vegetation index, green light normalized difference vegetation index, normalized water index, soil adjusted vegetation index, band mean and brightness. The characteristic value corresponding to each characteristic dimension is the average value of the corresponding features extracted from several sub-geographical area images.

[0026] Specifically, the normalized difference vegetation index NDVI = (Nir-Red) / (Nir+Red), where Nir is the reflectivity of the near-infrared band and Red is the reflectivity of the red light band.

[0027] Green normalized difference vegetation index GNDVI = (Nir-Green) / (Nir+Green), where Green is the reflectivity of the green light band.

[0028] Normalized water index NDWI=(Green-Nir) / (Green+Nir).

[0029] Soil Adjustment Vegetation Index SAVI=(1+L)×(Nir-Red) / (Nir+Red+L), where L is the soil adjustment coefficient, ranging from 0 to 1, and is adjusted according to the surface vegetation density.

[0030] As mentioned above, since different geographical areas may have crops, pond water bodies, wastelands, etc., it is necessary to obtain the vegetation, water and soil conditions on the surface of the geographical area. By obtaining the characteristic values ​​corresponding to the above-mentioned multiple characteristic dimensions, the distinction between land feature types can be more comprehensive and accurate, thereby improving the accuracy of geographical area type division.

[0031] S300, based on a number of geographic area samples obtained within a preset time period, clustering is performed on the feature vector samples corresponding to each geographic area sample, each obtained geographic area sample cluster is used as a sub-sample set to train a number of preset classification models, and a target classification model corresponding to each geographic area sample cluster is determined according to the geographic area type output result of each preset classification model.

[0032] Specifically, the geographic area samples refer to image samples obtained by segmenting remote sensing image samples according to a multi-scale segmentation model.

[0033] Furthermore, the feature vector samples corresponding to the geographic area samples are obtained in the same manner as the target feature vectors corresponding to the sub-geographical areas, which will not be described in detail herein.

[0034] Specifically, the feature vector samples corresponding to each geographic region sample may be clustered using an unsupervised clustering method, for example, hierarchical clustering, K-means clustering or fuzzy clustering, so as to cluster samples with similar feature vectors into the same cluster.

[0035] Furthermore, the geographical area type refers to a pre-set type of farmland subsidy issuance, which includes three types: subsidized plots, non-subsidized plots, and pending plots. In specific implementations, subsidized plots include plots where grain and oil crops, soybeans, potatoes, and vegetables are planted. Non-subsidized plots are mainly plots that have been changed to ponds and water bodies, construction land, and plots where perennial cash crops are planted, such as flowers and seedlings. Pending plots are mainly wasteland, simple Dapeng, and other plots. When obtaining geographical area samples, several geographical area samples are obtained for each type of plot.

[0036] In a specific embodiment, the target classification model corresponding to each geographical area sample cluster is determined by the following steps: S301, divide the several feature vector samples in each geographic area sample cluster into sub-training sets and sub-testing sets. Since the vegetation index of the same type of ground objects is roughly the same at the same shooting time, and the number of ground object types is small, the proportion of the training set can be appropriately reduced, and the proportion of the sub-training set and the sub-testing set can be set to 1:1.

[0037] S302, inputting a plurality of feature vector samples in a plurality of sub-training sets into a plurality of preset classification models, so that each preset classification model is learned according to the input plurality of feature vector samples and the geographical area type corresponding to each feature vector sample.

[0038] Preferably, there are at least two types of preset classification models, for example, a neural network model, a random forest prediction model, a logistic regression model, etc.

[0039] S303, inputting several feature vector samples in several sub-test sets into several learned preset classification models, and calculating the prediction accuracy score of each preset classification model for the geographic area sample cluster based on the difference between the predicted result of the geographic area type corresponding to each feature vector sample and the actual geographic area type corresponding to each feature vector sample.

[0040] Specifically, the prediction accuracy score of any preset classification model for the geographic region sample cluster meets the following conditions: F=∑ m i=1 (2T i / (2T i +N i +P i )) / m, where F represents the prediction accuracy score of the preset classification model for the geographical area sample cluster, T i N is the number of correct predictions among several feature vector samples predicted as the i-th geographic area type, i is the number of incorrect predictions in the feature vector samples predicted as the i-th geographic area type, P i is the number of feature vector samples that are not predicted as the i-th geographical area type but belong to the i-th geographical area type, m is the number of geographical area types, and the value range of i is 1 to m. In this embodiment, m=3.

[0041] In a specific embodiment, the random forest prediction model is used as an example to illustrate as follows: For all sub-training sets, several sub-training sets are randomly selected with replacement to form an input sample set, that is, the feature vector samples in the extracted several sub-training sets are used to form an input sample set, and the number of input sample sets is equal to the number of decision trees in the random forest prediction model. Each input sample set is separately input into a decision tree, so that the decision tree learns the input sample set, and the best feature is selected for splitting according to the information gain each time the split is performed until the training is completed; then for any sub-test set, several feature vector samples in the sub-test set are input into each decision tree model in the random forest prediction model, and the classification result of the geographical area type corresponding to each feature vector sample can be obtained according to any decision tree model, and voting is performed according to the output results of all decision tree models to obtain the final geographical area type corresponding to each feature vector sample, which is compared with the actual geographical area type corresponding to each feature vector sample, so as to obtain the above parameters, so as to calculate the prediction accuracy score of the random forest prediction model for the geographical area sample cluster.

[0042] S304: Using the preset classification model corresponding to the maximum prediction accuracy score as the target classification model corresponding to the geographic area sample cluster.

[0043] As mentioned above, since different preset classification models have different classification effects on different geographic area types, it is necessary to screen out a target classification model with good classification effect for each geographic area type. In the above process, the gap between the test results of each preset classification model and the actual geographic area type is taken into account, so that the screened target classification model is more suitable for the corresponding geographic area sample cluster, thereby improving the prediction accuracy in the subsequent geographic area type prediction.

[0044] S400 , determining a target sample cluster corresponding to each sub-geographical area from a plurality of geographic area sample clusters according to a target feature vector corresponding to each sub-geographical area.

[0045] Specifically, step S400 includes the following steps: S401, obtaining a central feature vector corresponding to the cluster center of each geographical area sample cluster.

[0046] S402, calculating the distance value between the target feature vector corresponding to the sub-geographical area and the central feature vector corresponding to the cluster center of each geographic area sample cluster.

[0047] Furthermore, a direct distance method is used to calculate the distance between the target feature vector and the central feature vector. Those skilled in the art are aware of the specific calculation process of calculating the sample distance in the cluster, which will not be described in detail here.

[0048] S403: Determine the geographical area sample cluster corresponding to the minimum distance value as the target sample cluster corresponding to the sub-geographical area.

[0049] S500: Input the target feature vector corresponding to any sub-geographical area into the target classification model corresponding to the target sample cluster, and determine the geographical area type of any sub-geographical area according to the output result of the target classification model.

[0050] As mentioned above, since the feature vector samples in each sub-geographical area sample cluster are relatively similar and are suitable for a certain preset classification model, by calculating the distance between the target feature vector and the central feature vector, the geographic area sample cluster that is most similar to the target feature vector can be found, thereby finding the preset classification model with the highest accuracy for it, thereby improving the accuracy of judging the geographic area type.

[0051] In a specific embodiment, the method further comprises the following steps: S10, based on the determined geographic area type of each sub-geographic area, select a number of sub-geographic areas corresponding to the subsidized plots as key geographic areas.

[0052] S20, for any selected key geographical area, divide the key geographical area into a number of regional grids of equal area, and obtain the target feature vector corresponding to each regional grid. For example, the side length of the regional grid is 3m.

[0053] S30, determining a target sample cluster corresponding to each regional grid from a plurality of geographical region sample clusters according to the target feature vector of each regional grid; S40, inputting the target feature vector corresponding to any regional grid into the target classification model corresponding to the target sample cluster, and determining the geographical area type of any regional grid according to the output result of the target classification model.

[0054] S50, selecting regional grids belonging to subsidized plots of land according to the geographical area type of each regional grid, and distributing cultivated land subsidies according to the total area of ​​the selected regional grids.

[0055] As mentioned above, since the crops in the subsidized plots are relatively concentrated and continuous plots, most of the plots can be well segmented, but some plots are mixed with small areas of other varieties, especially with unclear edges, and cannot be accurately identified and segmented when the multi-scale segmentation model is used. Therefore, it is necessary to further process the sub-geographical areas to achieve a more accurate estimate of the subsidized area.

[0056] In summary, the present invention provides a method for determining the type of a geographic region. First, each remote sensing image of a target geographic region is divided into a plurality of sub-geographic region images to obtain a plurality of sub-geographic region images corresponding to each sub-geographic region. Then, plot attribute features are extracted for each sub-geographic region image to obtain a target feature vector corresponding to the sub-geographic region. The target feature vector includes a plurality of feature dimensions that can be used to distinguish different geographic region types, so that the identified land object type is more comprehensive and accurate. Then, the feature vector samples corresponding to each geographic region sample are clustered, and each acquired geographic region sample cluster is used as a sub-sample set to train a plurality of preset classification models to determine the target classification model corresponding to each geographic region sample cluster. In this process, the gap between the test results of each preset classification model and the actual geographic region type is considered to make the screened target classification model more suitable for the corresponding geographic region sample cluster. According to the geographic region sample cluster that is most similar to the sub-geographic region, the geographic region type of the sub-geographic region is finally determined. Therefore, the present invention can screen out the classification model with the highest accuracy for the geographic region features, thereby improving the classification accuracy of the geographic region type.

[0057] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to implementing a method in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the geographic area type determination method provided in the above embodiment.

[0058] An embodiment of the present invention further provides an electronic device, comprising a processor and the aforementioned non-transitory computer-readable storage medium.

[0059] Although some specific embodiments of the present invention have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of protection of the present invention is defined by the appended claims.

Claims

1. A method for determining a geographical area type, characterized in that: The method comprises the following steps: Based on a plurality of remote sensing images of the target geographical area acquired within a preset time period, a multi-scale segmentation model is used to divide each remote sensing image of the target geographical area into a plurality of sub-geographical area images, so as to obtain a plurality of sub-geographical area images corresponding to each sub-geographical area; For a plurality of sub-geographical area images corresponding to each sub-geographical area, land attribute features are extracted from each sub-geographical area image to obtain a target feature vector corresponding to each sub-geographical area; Based on a number of geographic area samples obtained within a preset time period, clustering is performed on the feature vector samples corresponding to each geographic area sample, using each acquired geographic area sample cluster as a subsample set to train a number of preset classification models, and determining a target classification model corresponding to each geographic area sample cluster according to the geographic area type output result of each preset classification model; According to the target feature vector corresponding to each sub-geographical area, a target sample cluster corresponding to each sub-geographical area is determined from a plurality of geographic area sample clusters; The target feature vector corresponding to any sub-geographical area is input into the target classification model corresponding to the target sample cluster, and the geographical area type of any sub-geographical area is determined according to the output result of the target classification model.

2. The method for determining the type of geographical area according to claim 1, characterized in that: The following steps are used to obtain several sub-geographical area images corresponding to each sub-geographical area: Based on the number of sub-geographical area images corresponding to each remote sensing image, the number with the highest frequency of occurrence among several numbers is used as the target number corresponding to the sub-geographical area images, so as to screen out several remote sensing images corresponding to the target number; For the selected remote sensing images, the image center point coordinates of each sub-geographical area image in each remote sensing image are obtained, and all the image center point coordinates are clustered according to the target number to obtain the target number of image center point clusters; For any image center point cluster, the sub-geographic area images corresponding to the image center point coordinates whose distance from the cluster center of the image center point cluster is greater than a first preset distance threshold are deleted, and the sub-geographic area images corresponding to the remaining several image center point coordinates in the image center point cluster are taken as several sub-geographic area images of the same sub-geographic area.

3. The method for determining the type of geographical area according to claim 1, characterized in that: The characteristic dimensions corresponding to the target characteristic vector include normalized vegetation index, green light normalized difference vegetation index, normalized water index, soil adjusted vegetation index, band mean and brightness.

4. The method for determining the type of geographical area according to claim 1, characterized in that: Determine the target classification model corresponding to each geographic region sample cluster through the following steps: Divide a number of feature vector samples in each geographical area sample cluster into sub-training sets and sub-test sets; Inputting the plurality of feature vector samples in the plurality of sub-training sets into the plurality of preset classification models, so that each preset classification model is learned according to the plurality of input feature vector samples and the geographical area type corresponding to each feature vector sample; Inputting several feature vector samples in several subtest sets into several learned preset classification models, and calculating the prediction accuracy score of each preset classification model for the geographical area sample cluster according to the difference between the predicted result of the geographical area type corresponding to each feature vector sample and the actual geographical area type corresponding to each feature vector sample; The preset classification model corresponding to the maximum prediction accuracy score is used as the target classification model corresponding to the geographic area sample cluster.

5. The method for determining the type of geographical area according to claim 1, characterized in that: The step of determining a target sample cluster corresponding to each sub-geographical area from a plurality of geographic area sample clusters according to the target feature vector corresponding to each sub-geographical area comprises the following steps: Get the central feature vector corresponding to the cluster center of each geographic area sample cluster: Calculate the distance between the target feature vector corresponding to the sub-geographical area and the central feature vector corresponding to the cluster center of each geographic area sample cluster; The geographic area sample cluster corresponding to the minimum distance value is determined as the target sample cluster corresponding to the sub-geographic area.

6. The method for determining the type of geographical area according to claim 1, characterized in that: The geographical area type refers to a pre-set type of farmland subsidy issuance, and the farmland subsidy issuance type includes three types: subsidizable plots, non-subsidizable plots and pending plots.

7. The method for determining the type of geographic area according to claim 6, characterized in that: The method further comprises the steps of: According to the determined geographic area type of each sub-geographic area, several sub-geographic areas corresponding to the subsidized plots are selected as key geographic areas; For any selected key geographical area, the key geographical area is divided into a number of regional grids of equal area, and a target feature vector corresponding to each regional grid is obtained; According to the target feature vector of each regional grid, a target sample cluster corresponding to each regional grid is determined from a number of geographical region sample clusters; Inputting the target feature vector corresponding to any regional grid into the target classification model corresponding to the target sample cluster, and determining the geographical area type of any regional grid according to the output result of the target classification model; According to the geographical area type of each regional grid, the regional grids that belong to the subsidized plots are screened out, and the cultivated land subsidies are distributed according to the total area of ​​the screened regional grids.

8. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the method for determining the geographic area type as described in any one of claims 1-7.

9. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 8.

Images