A method, device and storage medium for mapping terraces based on remote sensing

Abstract

The application provides a remote sensing-based terrace mapping method, device, equipment and storage medium, wherein the method comprises the following steps: acquiring a land satellite remote sensing image and a digital elevation model of a target region; obtaining classification features of all pixels in the target region according to the land satellite remote sensing image and the digital elevation model, wherein the classification features are used to determine the land type corresponding to the pixels, and the land type comprises a terrace or a non-terrace; dividing the target region into multiple sub-regions, inputting the classification features of the pixels in each sub-region into a classifier corresponding to the sub-region, and obtaining the land type of the sub-region, wherein the classifier is obtained after training and testing using training samples and test samples corresponding to the sub-region; and comprehensively obtaining the classification results of each sub-region to obtain a terrace map of the target region, wherein the terrace map is in the form of pixels. The spatial distribution information of the terraces can be accurately and quickly obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of remote sensing mapping, in particular, to a remote sensing based terrace mapping method, device, equipment and storage medium. BACKGROUND

[0002] Terrace is an important farmland type, which has multiple ecological values, including reducing soil erosion, improving land productivity, increasing biodiversity, etc., therefore, accurate terrace spatial distribution information is of great significance for research and management.

[0003] Traditional land cover information acquisition relies on field investigation, however, this method is costly and inefficient, and it is difficult to quickly and accurately obtain large-scale terrace distribution information. Remote sensing technology can realize large-scale monitoring at low cost and high efficiency, and has been widely used in the field of land cover information extraction in recent years. At present, the research on terrace identification based on remote sensing technology mainly adopts visual interpretation, object-oriented classification, texture analysis and other methods, but these methods have low calculation efficiency, high dependence on research area for parameter setting, and poor universality; in addition, they rely on high-resolution satellite images or unmanned aerial images, which are costly and difficult to obtain on a large scale. The above limitations make it difficult to promote and apply these methods in large-scale research areas.

[0004] Therefore, in this field, there is an urgent need to develop an effective, efficient and universal terrace remote sensing mapping method to accurately and quickly obtain the spatial distribution information of terraces. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a remote sensing based terrace mapping method, device, equipment and storage medium to accurately and quickly obtain the spatial distribution information of terraces.

[0006] To achieve the above purpose, on the one hand, the embodiments of the present application provide a remote sensing based terrace mapping method, comprising:

[0007] obtaining land satellite remote sensing images and digital elevation models of a target area;

[0008] obtaining classification features of all pixels in the target area according to the land satellite remote sensing images and digital elevation models, the classification features being used to determine the land type corresponding to the pixels, wherein the land type includes terraces or non-terraces;

[0009] dividing the target area into multiple partitions, and inputting the classification features of the pixels in each partition into the classifier corresponding to the partition to obtain the land type of the partition, wherein the classifier is obtained by training and testing the training samples and test samples corresponding to the partition;

[0010] The classification results of each subregion are integrated to obtain a terrace map of the target region, the terrace map taking pixels as units.

[0011] Preferably, the classification features of all pixels in the target region obtained according to the Landsat remote sensing image and the digital elevation model further comprise:

[0012] The Landsat remote sensing image is preprocessed.

[0013] Time-series spectral features are extracted from the preprocessed Landsat remote sensing image.

[0014] The time-series spectral features are subjected to dimension reduction processing by extracting quantiles to obtain first features.

[0015] Spectral indices are calculated based on the time-series spectral features.

[0016] The spectral indices are subjected to dimension reduction processing by extracting quantiles to obtain second features.

[0017] Terrain information is extracted from the digital elevation model.

[0018] Terrain factors are calculated based on the terrain information to obtain third features, and the first features, the second features and the third features constitute the classification features of all pixels in the target region.

[0019] Preferably, the determination method of the training sample comprises:

[0020] Global samples are selected from the target region.

[0021] On the basis of the global samples, local samples corresponding to each subregion are selected for each subregion.

[0022] The global samples and the local samples of each subregion are integrated as the training sample corresponding to each subregion.

[0023] Preferably, the determination method of the test sample comprises:

[0024] The test sample is selected from the target region by using a hexagonal grid sampling strategy to ensure that the test sample is uniformly and randomly distributed in the target region.

[0025] Preferably, the test sample is selected from the target region by using the hexagonal grid sampling strategy further comprises:

[0026] The target region is divided into a plurality of hexagonal grids.

[0027] A plurality of first samples are randomly generated in each hexagonal grid.

[0028] determining, as a first selected grid, all hexagonal grids in which the first samples contain the terrace type;

[0029] determining, as a second selected grid, hexagonal grids adjacent to the first selected grid;

[0030] randomly generating a plurality of second samples in the first selected grid and the second selected grid;

[0031] determining, as a third selected grid, all hexagonal grids in which the first samples or the second samples contain the terrace type;

[0032] randomly generating a plurality of third samples in the third selected grid, the first samples, the second samples and the third samples being test samples of the target region.

[0033] Preferably, after the classification results of each sub-region are integrated to obtain the terrace map of the target region, the method further comprises:

[0034] calculating an accuracy value of the terrace map according to the test samples, to evaluate the accuracy of the terrace map;

[0035] calculating the uncertainty of each pixel in the target region according to the probabilities of different classification results of each pixel in the target region calculated by the classifier, to evaluate the uncertainty degree of the terrace map.

[0036] Preferably, the calculation of the uncertainty of each pixel in the target region according to the probabilities of different classification results of each pixel in the target region calculated by the classifier further comprises:

[0037] the uncertainty of each pixel in the target region is calculated by the following formula:

[0038] U = 1 - P max ;

[0039] wherein, U is the uncertainty of each pixel in the target region, P max is the larger one of a first probability value and a second probability value corresponding to each pixel, the first probability value being the probability value of each pixel being a terrace, and the second probability value being the probability value of each pixel being a non-terrace.

[0040] In another aspect, the embodiments of the present application provide a device for drawing a terrace map based on remote sensing, the device comprising:

[0041] an acquisition module configured to acquire a land satellite remote sensing image and a digital elevation model of a target region;

[0042] The classification feature determination module is configured to obtain classification features of all pixels in the target region according to the land satellite remote sensing image and the digital elevation model, and the classification features are used to determine a land type corresponding to the pixels, wherein the land type includes terraced fields or non-terraced fields.

[0043] The classification module is configured to divide the target region into a plurality of sub-regions, input the classification features of the pixels in each sub-region into a classifier corresponding to the sub-region, and obtain a land type of the sub-region, wherein the classifier is obtained by training and testing using training samples and testing samples corresponding to the sub-region.

[0044] The synthesis module is configured to synthesize the classification results of each sub-region to obtain a terraced field map of the target region, and the terraced field map is in units of pixels.

[0045] In another aspect, the embodiments herein also provide a computer device comprising a memory, a processor, and a computer program stored on the memory, which, when executed by the processor, performs the instructions of any of the methods described above.

[0046] In another aspect, the embodiments herein also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor of a computer device, performs the instructions of any of the methods described above.

[0047] As can be seen from the technical solutions provided by the embodiments herein, the embodiments herein train and test the classifier corresponding to each sub-region using samples, on the one hand, after the target region is divided into a plurality of sub-regions, each sub-region is a small range relative to the target region, and under the same precision of the classifier classification, the number of training samples required by the small range is relatively small, so that the training efficiency of the classifier is not low due to the excessive number of samples of each sub-region, and the total amount of sampling work can be reduced by sharing part of the training samples in each sub-region; on the other hand, each sub-region has its corresponding classifier, and the classifier is trained and tested by the samples in each sub-region, so that the classifier is more suitable for the corresponding sub-region, and the classification effect is better and the precision is higher when the classifier is used for classification.

[0048] In order to make the above and other purposes, features and advantages of the present embodiments more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments herein or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments herein, and other drawings can be obtained by those skilled in the art without any creative effort.

[0050] Figure 1 A flowchart of a remote sensing-based terrace mapping method provided by the embodiments is shown;

[0051] Figure 2 A flowchart of obtaining classification features of all pixels in a target region according to Landsat remote sensing images and a digital elevation model is shown;

[0052] Figure 3 A flowchart of determining training samples is shown;

[0053] Figure 4 A flowchart of selecting test samples from a target region by using a hexagonal grid sampling strategy is shown;

[0054] Figure 5 A flowchart of further evaluating the accuracy and uncertainty degree of a terrace map after synthesizing classification results of each subregion to obtain the terrace map of the target region is shown;

[0055] Figure 6 A module structure diagram of a remote sensing-based terrace mapping device provided by the embodiments is shown;

[0056] Figure 7 A structure diagram of a computer device provided by the embodiments is shown.

[0057] List of Symbols:

[0058] 100, an acquisition module;

[0059] 200, a classification feature determination module;

[0060] 300, a classification module;

[0061] 400, a synthesis module;

[0062] 702, a computer device;

[0063] 704, a processor;

[0064] 706, a memory;

[0065] 708, a driving mechanism;

[0066] 710, an input / output module;

[0067] 712, an input device;

[0068] 714, an output device;

[0069] 716, a presentation device;

[0070] 718, graphical user interface;

[0071] 720, network interface;

[0072] 722, communication link;

[0073] 724, communication bus. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments herein, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0075] Traditional land cover information acquisition relies on field investigation, however, this method is high in cost and low in efficiency, and it is difficult to quickly and accurately obtain large-scale terrace distribution information. Remote sensing technology can realize large-scale monitoring at low cost and high efficiency, and has been widely used in the field of land cover information extraction in recent years. At present, the research on terrace identification based on remote sensing technology mainly adopts visual interpretation, object-oriented classification, texture analysis and other methods, but these methods are low in calculation efficiency, highly dependent on the research area in parameter setting, and poor in universality; in addition, they depend on high-resolution satellite images or unmanned aerial images, which are high in data use cost and difficult to obtain in a large range. The above limitations make it difficult to popularize and apply these methods in a large research area.

[0076] To solve the above problems, the embodiments of the present application provide a remote sensing-based terrace mapping method. Figure 1 It is a flowchart of a remote sensing-based terrace mapping method provided by the embodiments of the present application, and the present specification provides method operation steps as described in the embodiments or flowchart, but more or fewer operation steps can be included based on conventional or non-creative work. The order of steps listed in the embodiments is only one of the many execution orders of steps, and does not represent the only execution order. When the system or device product is executed in practice, it can be executed in sequence or in parallel according to the method order shown in the embodiments or drawings.

[0077] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting, unless the context clearly indicates otherwise. For example, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof. As used herein, the term "if' can be construed to mean "when" or "when not" that a condition is met or not met, unless otherwise indicated.

[0078] Referring to Figure 1 A remote sensing based terrace mapping method is provided herein, comprising:

[0079] S101: obtaining a land satellite remote sensing image and a digital elevation model of a target region;

[0080] S102: obtaining classification features of all pixels in the target region according to the land satellite remote sensing image and the digital elevation model, the classification features being used to determine the land type corresponding to the pixels, wherein the land type comprises a terrace or a non-terrace;

[0081] S103: dividing the target region into multiple sub-regions, inputting the classification features of the pixels in each sub-region into a classifier corresponding to the sub-region, and obtaining the land type of the sub-region, wherein the classifier is obtained by training and testing using training samples and testing samples corresponding to the sub-region;

[0082] S104: synthesizing the classification results of each sub-region to obtain a terrace map of the target region, the terrace map being in units of pixels.

[0083] The target region is a region that needs to be mapped for terraces, and the land satellite remote sensing image and the digital elevation model can be obtained according to information periodically released by an authoritative agency. Referring to Figure 2 Further, the step of obtaining classification features of all pixels in the target region according to the land satellite remote sensing image and the digital elevation model further comprises:

[0084] S201: preprocessing the land satellite remote sensing image;

[0085] S202: extracting time series spectral features from the preprocessed land satellite remote sensing image;

[0086] S203: performing dimensionality reduction processing on the time series spectral feature extraction quantile to obtain a first feature;

[0087] S204: calculate a spectral index based on the time-series spectral features;

[0088] S205: perform dimensionality reduction processing on the extracted quantiles of the spectral index to obtain a second feature;

[0089] S206: extract terrain information from the digital elevation model;

[0090] S207: calculate a terrain factor based on the terrain information to obtain a third feature, and the first feature, the second feature, and the third feature form the classification features of all pixels in the target region.

[0091] For land satellite remote sensing images, cloud mask preprocessing is needed to prevent remote sensing images from being affected by clouds and to reduce the accuracy of subsequent terrace mapping. Further, time-series spectral features are extracted from the preprocessed land satellite remote sensing images. Specifically, time-series spectral features are often used to describe the phenological information of agricultural land cover. However, to avoid the Hughes phenomenon caused by too many time-series spectral features, dimensionality reduction processing is needed on the extracted quantiles of the time-series spectral features. The time-series spectral features are the time-series spectral features of the year to be classified.

[0092] In the embodiments of the present application, the time-series spectral features can include six optical bands and one thermal infrared band. The six optical bands include: blue band (Blue), green band (Green), red band (Red), near-infrared band (NIR), short-wave infrared 1 band (SWIR1), and short-wave infrared 2 band (SWIR2). The thermal infrared band is thermal infrared 2 band (TIRS2). The extracted quantiles can be the quartiles of the six optical bands and one thermal infrared band, and the first feature is obtained.

[0093] Based on the time-series spectral features, a spectral index can be calculated. In the embodiments of the present application, the spectral index can include: normalized difference vegetation index (NDVI), improved normalized difference water index (MNDWI), normalized difference building index (NDBI), and bare soil index (BSI). Similarly, to avoid the Hughes phenomenon caused by too many spectral indices, dimensionality reduction processing is needed on the extracted quantiles of the spectral indices, and in the embodiments of the present application, the quartiles can be extracted.

[0094] Specifically, the calculation formula of the normalized difference vegetation index is:

[0095]

[0096] wherein, NDVI is the normalized difference vegetation index, NIR is the near-infrared band, and Red is the red band.

[0097] The calculation formula of the improved normalized difference water index is:

[0098]

[0099] wherein, MNDWI is a modified normalized difference water index, Green is a green band, and SWIR1 is a short wave infrared 1 band.

[0100] The calculation formula of the normalized difference building index is:

[0101]

[0102] wherein, NDBI is a normalized difference building index, SWIR1 is a short wave infrared 1 band, and NIR is a near infrared band.

[0103] The calculation formula of the bare soil index is:

[0104]

[0105] wherein, BSI is a bare soil index, SWIR1 is a short wave infrared 1 band, Red is a red band, Blue is a blue band, and NIR is a near infrared band.

[0106] In addition, terrain information can also be extracted based on a digital elevation model, and in the embodiments of the present document, the terrain information can include: an elevation variation, a horizontal distance, an elevation, an average value of elevations within a certain range, a maximum value of elevations within a certain range, and a minimum value of elevations within a certain range, all of which can be obtained through the digital elevation model, wherein the certain range can be determined according to actual working conditions, and the present document does not make specific limitations.

[0107] A terrain factor is calculated based on the terrain information, that is, a third feature for describing terrain characteristics is obtained, and the third feature includes: an elevation (H), a slope (S), a slope rate of change (SOS), a roughness (R), a slope shape (P), and a relief degree (RF).

[0108] Specifically, the calculation formula of the slope is:

[0109]

[0110] wherein, S is the slope, ΔH is the elevation variation, and ΔL is the horizontal distance.

[0111] The calculation formula of the slope rate of change is:

[0112]

[0113] wherein, SOS is the slope rate of change, ΔS is the slope variation, which can be extracted from the calculated slope data, and ΔL is the horizontal distance.

[0114] The calculation formula of the roughness is:

[0115]

[0116] wherein R is roughness, and S is slope.

[0117] The calculation formula of slope is:

[0118] P = H - H mean

[0119] wherein P is slope, H is elevation, and H mean is the average value of the elevation within a certain range.

[0120] The calculation formula of relief is:

[0121] RF = H max - H min wherein RF is relief, H max is the maximum value of the elevation within a certain range, and H min is the minimum value of the elevation within a certain range.

[0122] The first feature, the second feature, and the third feature described above constitute the classification features of all pixels in the target region, that is, one pixel corresponds to one classification feature, and one classification feature includes the first feature, the second feature, and the third feature.

[0123] In the prior art, a plurality of samples are generally collected in the target region, and then a classifier is trained and tested through the samples, and then the land type of the target region is identified as terraces or non-terraces by using the pre-trained classifier. The prior art scheme has the following problems: In order to ensure the accuracy of the classification of the classifier, a large number of training samples need to be collected, which increases the sampling workload, and a large number of samples input into the classifier leads to low training efficiency of the classifier; if the number of samples is reduced in order to improve work efficiency, the classification accuracy of the classifier will be reduced. Therefore, in the embodiments of the present text, the target region is divided into a plurality of sub-regions, and part of the training samples are shared in each sub-region, and the specific sub-regional division method can be equal division or division according to a certain attribute, for example, if the target region is the whole country, the regional range of a province / municipality directly under the central government can be taken as a sub-region, and if the target is a province, the regional range of a city at the next level under the province can be taken as a sub-region.

[0124] For each partition, each partition has a corresponding classifier. After the target area is divided into multiple partitions, each partition can be sampled respectively, and the classifier corresponding to each partition can be trained and tested by using the samples, thereby solving the problem that the classification accuracy and work efficiency of the classifier cannot be effectively balanced in the prior art. On the one hand, after the target area is divided into multiple partitions, each partition is a small range relative to the target area, and the number of samples required by the small range is relatively small under the same premise of the classification accuracy of the classifier. Therefore, the training efficiency of the classifier will not be low due to the excessively large number of samples of each partition, and the total amount of sampling work can be reduced by sharing part of the training samples in each partition. On the other hand, as analyzed above, each partition has a corresponding classifier. After the classifier is trained and tested by using the samples in each partition, the classifier is more suitable for the corresponding partition, and the classification effect is better and the accuracy is higher when the classifier is used for classification.

[0125] In the embodiments of the present document, reference is made to Figure 3 The determination method of the training sample includes:

[0126] S301: selecting a global sample from the target area;

[0127] S302: on the basis of the global sample, selecting a local sample corresponding to each partition for each partition;

[0128] S303: integrating the global sample and the local sample of each partition as the training sample corresponding to each partition.

[0129] The global sample selected from the target area is obtained by sampling the target area globally. On the basis of the global sample, the local sample is selected for each partition. The local sample is obtained by sampling the partition. When the local sample is sampled, it should be avoided to repeat the global sample. For each partition, the training sample corresponding to the partition includes the global sample and the local sample of the partition. In this way, the global sample can be collected to obtain the samples with universality in the target area, and then the local sample can be collected to obtain the samples specific to each partition. The global sample can be used for the training of the classifier of each partition, thereby reducing the total amount of work for collecting the training sample. The local sample plays a role in supplementing and enriching the global sample, and the two together constitute the training sample of each partition to train the classifier.

[0130] It should be noted that one global sample or local sample mentioned in the embodiments of the present document corresponds to one pixel. The type of the global sample or the local sample is the land type corresponding to the pixel, and the land type includes terraced fields or non-terraced fields. The type of the global sample or the local sample can be determined by visual method, and of course can be determined by other methods, which are not limited in the present document.

[0131] When training the classifier by using the training samples, the classifier can be trained by using the global samples first, and then the classifier is applied to different partitions to verify the classification accuracy of the classifier. If the classification accuracy is high, the corresponding partition does not need to be supplemented with local samples. If the classification accuracy is low, the corresponding partition is supplemented with local samples, and the classifier is trained by using the global samples and the local samples corresponding to the partition. When evaluating the classification accuracy, a certain amount of test samples can be input into the classifier according to the test samples, and the classification results of the classifier are compared with the actual results. The proportion of test samples in which the classification results are consistent with the actual results in all test samples is calculated. If the proportion is greater than a set threshold, it is proved that the classification accuracy is high, and vice versa.

[0132] In the embodiments of the present application, the method for determining the test samples comprises:

[0133] The test samples are selected from the target region by using a hexagonal grid sampling strategy, so as to ensure that the test samples are uniformly and randomly distributed in the target region.

[0134] Specifically, with reference to Figure 4 , the method for selecting the test samples from the target region by using the hexagonal grid sampling strategy further comprises:

[0135] S401: dividing the target region into a plurality of hexagonal grids;

[0136] S402: randomly generating a plurality of first samples in each hexagonal grid;

[0137] S403: determining a hexagonal grid in which the first samples contain terraced fields as a first selected grid;

[0138] S404: determining a hexagonal grid adjacent to the first selected grid as a second selected grid;

[0139] S405: randomly generating a plurality of second samples in the first selected grid and the second selected grid;

[0140] S406: determining a grid in which the first samples or the second samples contain terraced fields as a third selected grid;

[0141] S407: randomly generating a plurality of third samples in the third selected grid, and the first samples, the second samples and the third samples are used as the test samples of the target region.

[0142] The test samples obtained by using the hexagonal grid sampling strategy can be used to test the classifier of all partitions. In the prior art, the test samples are mostly randomly selected. However, since the proportion of the terrace area in the total area of the target region is usually small, random selection will result in fewer test samples of the terrace type in the target region, and the sample distribution is uneven. Therefore, in order to ensure the number and uniformity of the test samples, the test samples can be selected by the above steps S401-S407.

[0143] The second selected grid is a hexagonal grid adjacent to the first selected grid, that is, each first selected grid has six adjacent second selected grids. The grid in which the first sample or the second sample contains the terrace type is a third selected grid. The purpose of each step is to increase the number of test samples of the terrace type as much as possible. In the embodiment, one test sample corresponds to one pixel, and the type of the test sample is the land type corresponding to the pixel. The land type includes a terrace or a non-terrace. The type of the test sample can be determined by visual method, and can also be determined by other methods, which is not limited in the present embodiment.

[0144] After the classifier is trained by using the training samples of different partitions, the classifier can identify the land type of the corresponding partition. The classification features of all pixels in the partition are input into the classifier, and the classifier can output the land type corresponding to all pixels, that is, the terrace map of the partition can be determined. It should be noted that, in the embodiment, the obtained farmland range data can be used to mask the non-farmland region to improve the classification efficiency. Finally, all partitions are integrated to obtain the terrace map of the target region.

[0145] In pixel-based image classification, different objects have the same spectrum and the same object has different spectra, which will cause the salt and pepper effect. Since the shape of the terrace is irregular and the patch is fragmented, the salt and pepper effect in classification is particularly serious. Therefore, a filter parameter is selected to perform a mode filtering process on the classification result to eliminate the salt and pepper effect. In order to further improve the accuracy, after the mode filtering, small patch filtering is performed on the classification result, and a threshold is set to filter out small patch terraces and small patch non-terraces in turn.

[0146] In the embodiment, the reference Figure 5 The accuracy and uncertainty of the terrace map can be further evaluated. Specifically, after the classification results of each partition are integrated to obtain the terrace map of the target region, the method further comprises:

[0147] S501: calculating the accuracy value of the terrace map according to the test samples, to evaluate the accuracy of the terrace map;

[0148] S502: Calculate the uncertainty of each pixel in the target region according to the probability of each pixel corresponding to different classification results calculated by the classifier, so as to evaluate the uncertainty degree of the terrace map.

[0149] The calculation of the uncertainty of each pixel in the target region according to the probability of each pixel corresponding to different classification results calculated by the classifier further comprises:

[0150] The uncertainty of each pixel in the target region is calculated by the following formula:

[0151] U = 1-P max

[0152] Wherein, U is the uncertainty of each pixel in the target region, P max is the larger value of the first probability value and the second probability value corresponding to each pixel, the first probability value is the probability value of each pixel being a terrace, and the second probability value is the probability value of each pixel being a non-terrace.

[0153] When the classifier outputs the classification results corresponding to each pixel, it generally reflects the probability of each pixel corresponding to different land types, for example, the probability of pixel A being a terrace is 40%, and the probability of being a non-terrace is 60%, and the output classification result is a non-terrace.

[0154] According to the larger value of the probability value of each pixel corresponding to the terrace and the probability value of the non-terrace, the uncertainty of each pixel can be calculated, and then the uncertainty of each pixel can be calculated. The uncertainty degree of the terrace map can be evaluated. For example, the average of all uncertainties in the target region can be calculated, and if the average is greater than a set threshold, it means that the uncertainty degree of the terrace map is high.

[0155] In addition, the accuracy value of the terrace map can also be calculated according to the test sample to evaluate the accuracy of the terrace map. Specifically, the overall accuracy, user accuracy, producer accuracy and kappa coefficient of the terrace map can be calculated:

[0156] The calculation formula of the overall accuracy is:

[0157]

[0158] The calculation formula of the user accuracy is:

[0159]

[0160]

[0161] The calculation formula of the producer accuracy is:

[0162]

[0163]

[0164] The calculation formula of the kappa coefficient is:

[0165]

[0166]

[0167] Wherein, OA is the overall accuracy, UA0 is the user accuracy of the non-terraced type, UA1 is the user accuracy of the terraced type, PA0 is the producer accuracy of the non-terraced type, PA1 is the producer accuracy of the terraced type, and kappa is the kappa coefficient. 00 N is the number of samples in the test sample that are actually non-terraced and classified as non-terraced by the classifier, N 11 N is the number of samples in the test sample that are actually terraced and classified as terraced by the classifier, N 01 N is the number of samples in the test sample that are actually non-terraced and classified as terraced by the classifier, N 10 N is the number of samples in the test sample that are actually terraced and classified as non-terraced by the classifier.

[0168] The calculation formula of the overall accuracy, the user accuracy, the producer accuracy, and the kappa coefficient described above can be calculated based on the test sample in the target area, or can be calculated based on the test sample in the subzone. It is not limited in this paper whether the test sample in which range is calculated. The above calculation formula can calculate the overall accuracy, the user accuracy, the producer accuracy, and the kappa coefficient in the corresponding range. Generally, the higher the overall accuracy, the user accuracy, the producer accuracy, and the kappa coefficient, the higher the accuracy of the terraced mapping.

[0169] Based on the above-mentioned method for mapping terraced fields based on remote sensing, the embodiments of the present application also provide a device for mapping terraced fields based on remote sensing. The device can include a system (including a distributed system), software (application), module, component, server, client, etc. using the method described in the embodiments of the present application, and a device combined with necessary implementation hardware. Based on the same innovative concept, the device in one or more embodiments provided by the embodiments of the present application is described as follows. Since the implementation scheme of the device for solving the problem is similar to the method, the implementation of the specific device in the embodiments of the present application can be referred to the implementation of the foregoing method, and the repeated parts will not be described herein. The term "unit" or "module" used below can be a combination of software and / or hardware that can realize a predetermined function. Although the device described in the following embodiments is preferably realized in software, the realization of hardware or a combination of software and hardware is also possible and is conceived.

[0170] Specifically, Figure 6is a schematic diagram of a module structure of one embodiment of a terrace mapping device based on remote sensing provided in the embodiments of the present application, referring to Figure 6 The terrace mapping device based on remote sensing provided in the embodiments of the present application includes: an acquisition module 100, a classification feature determination module 200, a classification module 300, and a comprehensive module 400.

[0171] The acquisition module 100 is configured to acquire a land satellite remote sensing image and a digital elevation model of a target region.

[0172] The classification feature determination module 200 is configured to obtain classification features of all pixels in the target region according to the land satellite remote sensing image and the digital elevation model, the classification features being used to determine land types corresponding to the pixels, wherein the land types include terraces or non-terraces.

[0173] The classification module 300 is configured to divide the target region into multiple sub-regions, input the classification features of the pixels in each sub-region into a classifier corresponding to the sub-region, and obtain a land type of the sub-region, wherein the classifier is obtained by training and testing using training samples and testing samples corresponding to the sub-region.

[0174] The comprehensive module 400 is configured to comprehensively analyze classification results of each sub-region, and obtain a terrace map of the target region, the terrace map being in units of pixels.

[0175] Referring to Figure 7As shown, based on the above-described remote sensing based terrace mapping method, one embodiment of the present disclosure further provides a computer device 702, wherein the above-described method is run on the computer device 702. The computer device 702 can include one or more processors 704, such as one or more central processing units (CPUs) or graphics processing units (GPUs), each of which can implement one or more hardware threads. The computer device 702 can also include any memory 706 for storing any kind of information, such as code, settings, data, etc., in one implementation, a computer program stored on the memory 706 and executable on the processor 704, when executed by the processor 704, can perform instructions according to the above-described method. Without limitation, for example, the memory 706 can include any one or combination of the following: any type of RAM, any type of ROM, a flash memory device, a hard disk, an optical disk, etc. More generally, any memory can store information using any technology. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory can represent a fixed or removable component of the computer device 702. In one case, the computer device 702 can perform any operation of the associated instructions when the processor 704 executes the associated instructions stored in any memory or combination of memories. The computer device 702 also includes one or more drive mechanisms 708, such as a hard disk drive mechanism, an optical disk drive mechanism, etc., for interacting with any memory.

[0176] The computer device 702 can also include an input / output module 710 (I / O) for receiving various inputs (via input devices 712) and for providing various outputs (via output devices 714). One particular output mechanism can include a presentation device 716 and an associated graphical user interface 718 (GUI). In other embodiments, the input / output module 710 (I / O), the input devices 712, and the output devices 714 can also not be included, just as a computer device in a network. The computer device 702 can also include one or more network interfaces 720 for exchanging data with other devices via one or more communication links 722. One or more communication buses 724 couple the above-described components together.

[0177] The communication links 722 can be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication links 722 can include any combination of hardwired links, wireless links, routers, gateway functionality, name servers, etc., governed by any protocol or combination of protocols.

[0178] Corresponding to Figures 1-5In addition to the methods described above, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the above-described methods.

[0179] This embodiment also provides a computer-readable instruction, wherein when a processor executes the instruction, the program therein causes the processor to perform the following: Figures 1 to 5 The method shown.

[0180] It should be understood that in the various embodiments of this document, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this document.

[0181] It should also be understood that, in the embodiments herein, the term "and / or" is merely a description of the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following associated objects have an "or" relationship.

[0182] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this document.

[0183] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0184] In the embodiments provided herein, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, devices, or units, or they may be electrical, mechanical, or other forms of connection.

[0185] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e. may be located in one place, or may be distributed to multiple network units. Part or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0186] In addition, each functional unit in each embodiment herein can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0187] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions herein, the essential part or the whole or part of the prior art, or the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment herein. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0188] The principles and implementation modes of the present application are described in the specific embodiments herein, and the above embodiment descriptions are only used to help understand the method and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed; in view of the above, the content of the present application should not be understood as a limitation.

Claims

1. A method for terracing based on remote sensing, characterized in that, include: Acquire land satellite remote sensing images and digital elevation models of the target area; Based on the land satellite remote sensing image and digital elevation model, the classification features of all pixels in the target area are obtained. The classification features are used to determine the land type corresponding to the pixel, wherein the land type includes terraced fields or non-terraced fields. The target area is divided into multiple partitions. The classification features of the pixels in each partition are input into the classifier corresponding to that partition to obtain the land type of that partition. The classifier is obtained by training and testing using the training samples and test samples corresponding to that partition. By combining the classification results of each partition, a terraced field map of the target area is obtained, wherein the terraced field map is in pixels; The method for determining the training samples includes: Select global samples from the target region; Based on the global samples, local samples of each partition are selected for each partition. The global samples and the local samples of each partition are combined as the training samples for each partition. When training a classifier using training samples, you can first train the classifier using global samples, and then apply the classifier to different partitions to verify the classification accuracy. If the classification accuracy is high, the corresponding partition does not need to be supplemented with local samples. If the classification accuracy is low, local samples should be supplemented in the corresponding partition.

2. The method for terracing based on remote sensing according to claim 1, characterized in that, The step of obtaining the classification features of all pixels in the target area based on the land satellite remote sensing image and digital elevation model further includes: The land satellite remote sensing images are preprocessed; Extracting temporal spectral features from preprocessed Landsat remote sensing images; The quantiles of the extracted time-series spectral features are subjected to dimensionality reduction processing to obtain the first feature; Based on the aforementioned time-series spectral characteristics, the spectral index is calculated; The quantiles of the spectral index are extracted and then subjected to dimensionality reduction processing to obtain the second feature; Extract terrain information from the digital elevation model; Based on the terrain information, a terrain factor is calculated to obtain a third feature. The first feature, the second feature, and the third feature constitute the classification features of all pixels in the target area.

3. The method for terracing based on remote sensing according to claim 1, characterized in that, The method for determining the test sample includes: A hexagonal grid sampling strategy is used to select test samples from the target area to ensure that the test samples are uniformly and randomly distributed in the target area.

4. The method for terracing based on remote sensing according to claim 3, characterized in that, The method of selecting test samples from the target region using a hexagonal grid sampling strategy further includes: The target area is divided into several hexagonal grids; Several first samples are randomly generated in each hexagonal grid; The first hexagonal grid in all hexagonal grids that contains a terraced field type is identified as the first selected grid. The hexagonal grid adjacent to the first selected grid is designated as the second selected grid. Several second samples are randomly generated in the first and second selected grids; The grids in which the first or second sample of all hexagonal grids contains the terrace type are identified as the third selected grids; Several third samples are randomly generated in the third selected grid, and the first, second and third samples are used as test samples for the target area.

5. The method for terracing based on remote sensing according to claim 1, characterized in that, After synthesizing the classification results of each partition to obtain the terraced field map of the target area, the process also includes: Based on the test samples, the accuracy value of the terraced field map is calculated to evaluate the accuracy of the terraced field map; Based on the probability of each pixel in the target area corresponding to different classification results calculated by the classifier, the uncertainty of each pixel in the target area is calculated to evaluate the degree of uncertainty of the terraced field map.

6. The method for terracing based on remote sensing according to claim 5, characterized in that, The calculation of the uncertainty of each pixel in the target region, based on the probability of different classification results for each pixel calculated by the classifier, further includes: The uncertainty of each pixel in the target region is calculated using the following formula: U=1-P max ; Where U represents the uncertainty of each pixel in the target region, and P max The value is the larger of the first probability value and the second probability value for each pixel. The first probability value is the probability that each pixel is a terraced field, and the second probability value is the probability that each pixel is not a terraced field.

7. A remote sensing-based terraced field mapping device, characterized in that, The device includes: The acquisition module is used to acquire land satellite remote sensing images and digital elevation models of the target area; The classification feature determination module is used to obtain the classification features of all pixels in the target area based on the land satellite remote sensing image and digital elevation model. The classification features are used to determine the land type corresponding to the pixel, wherein the land type includes terraced fields or non-terraced fields. The classification module is used to divide the target area into multiple partitions, input the classification features of the pixels in each partition into the classifier corresponding to that partition, and obtain the land type of that partition. The classifier is obtained by training and testing using training samples and test samples corresponding to that partition. The integration module is used to integrate the classification results of each partition to obtain a terraced field map of the target area, wherein the terraced field map is in pixels; The method for determining the training samples includes: Select global samples from the target region; Based on the global samples, local samples of each partition are selected for each partition. The global samples and the local samples of each partition are combined as the training samples for each partition. When training a classifier using training samples, you can first train the classifier using global samples, and then apply the classifier to different partitions to verify the classification accuracy. If the classification accuracy is high, the corresponding partition does not need to be supplemented with local samples. If the classification accuracy is low, local samples should be supplemented in the corresponding partition.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the computer program is run by the processor, it executes the instructions of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor of the computer device, it executes the instructions of the method according to any one of claims 1-6.

Images