CN116206197B - Ground science partitioning method for farmland information extraction - Google Patents
Ground science partitioning method for farmland information extraction Download PDFInfo
- Publication number
- CN116206197B CN116206197B CN202211627162.9A CN202211627162A CN116206197B CN 116206197 B CN116206197 B CN 116206197B CN 202211627162 A CN202211627162 A CN 202211627162A CN 116206197 B CN116206197 B CN 116206197B
- Authority
- CN
- China
- Prior art keywords
- cultivated land
- cultivated
- land
- area
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000000605 extraction Methods 0.000 title claims abstract description 28
- 238000000638 solvent extraction Methods 0.000 title claims abstract description 26
- 238000009826 distribution Methods 0.000 claims abstract description 88
- 230000000877 morphologic effect Effects 0.000 claims abstract description 6
- 238000006243 chemical reaction Methods 0.000 claims abstract description 4
- 238000005192 partition Methods 0.000 claims description 33
- 230000011218 segmentation Effects 0.000 claims description 19
- 238000012876 topography Methods 0.000 claims description 18
- 206010035148 Plague Diseases 0.000 claims description 12
- 241000607479 Yersinia pestis Species 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 238000013145 classification model Methods 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 4
- 238000003064 k means clustering Methods 0.000 claims description 4
- 238000007637 random forest analysis Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 3
- 238000013316 zoning Methods 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 2
- 238000003786 synthesis reaction Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 5
- 238000004458 analytical method Methods 0.000 abstract description 2
- 238000013507 mapping Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 238000003709 image segmentation Methods 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 239000002689 soil Substances 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000002194 synthesizing effect Effects 0.000 description 2
- 230000000699 topical effect Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 244000068988 Glycine max Species 0.000 description 1
- 235000010469 Glycine max Nutrition 0.000 description 1
- 240000008042 Zea mays Species 0.000 description 1
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 description 1
- 235000002017 Zea mays subsp mays Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 235000005822 corn Nutrition 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010348 incorporation Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000013439 planning Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a method for distinguishing a ground science oriented to tilling information extraction, which is characterized by comprising the following steps of: 1) Establishing a cultivated land type interpretation mark and a classification system; 2) Establishing a regular honeycomb grid, and collecting samples which are uniformly distributed in the grid in space; 3) Combining with multi-source data, constructing a geoscience partitioning method by means of clustering, morphological processing, spatial data superposition analysis and the like, and realizing multi-level geoscience partitioning oriented to interpretation requirements; 4) Dividing the cultivated land information extraction area into 4 sub-areas; 5) Performing object-to-point-to-object sample conversion on the sample obtained in the step 2; 6) And in each subarea, the tilling information extraction under the support of the geography subarea is realized. The geologic partitioning method comprehensively utilizes the thought of geologic partitioning, combines multi-source data, and defines the centralized distribution areas of different forms of cultivated lands by the partitioning method so as to realize high-precision automatic extraction of different cultivated land types.
Description
Technical Field
The invention belongs to the technical field of digital image processing, and particularly relates to a geologic partitioning method for extracting remote sensing information of cultivated land.
Background
Grain safety is a major challenge for human sustainable development. As an important data source for agricultural mapping and monitoring, a great deal of research has been conducted on information extraction of cultivated lands and crops based on remote sensing, and the information mainly comprises information such as crop distribution, crop types, planting frequency and the like, and the research proves the feasibility and application prospect of the remote sensing technology applied to the information extraction of the cultivated lands. With the improvement of the resolution of the images, the visual features of the cultivated land in the images are gradually thinned. The cultivated land has different expression forms according to different geographic environments, for example, the cultivated land in a land area with a flat topography is mostly represented as a square or rectangular land block with a regular boundary, the cultivated land at the junction of the mountain land and the plain is mostly accompanied by planting textures but has no regular boundary, and the cultivated land in a mountain area with relatively large relief is mostly represented as a strip shape with a spatial up-down structure. Although the distribution area of the cultivated land at the hillside is not large compared to the cultivated land in a gentle region of topography, they play an important role. Such as terraces, are constructed to convert mountain areas or steep slopes into arable land, and many studies have demonstrated the importance of terraces as an integrated system of land and water. In addition, the hillside cultivated land is used as an important component of hillside cultivated land, and has weak capability of resisting drought, flood and other natural disasters due to extensive cultivation, so that the hillside cultivated land has important significance in the aspects of reducing water and soil loss, reasonably planning and utilizing the hillside cultivated land resources, compensating the occupied cultivated land for industrial and town development and the like. Therefore, the space distribution of cultivated lands with different forms is significant for the ecological environment management of cultivated lands and the national grain safety assurance (See, fritz S, see L, mccall I, et al mapping global cropland and field size [ J ]. Global change biology,2015,21 (5): 1980-1992.).
Current research on remote sensing tilling information extraction is mostly focused on identification of tilled, non-tilled, or specific crop types, such as corn, soybean, etc., and relatively few tilling extraction researches for different visual features. The cultivated lands with different forms have the following two main difficulties in extraction: firstly, although we recognize that the morphology of cultivated lands is different, the image features displayed on the remote sensing image due to the problems of 'homography' or 'foreign matter homography' cannot be accurately distinguished, although there are many ways of assisting in recognition by means of the time sequence growth features of crops (see Cao B, yu L, naipal V, et al.a30m terrace mapping in China using Landsat 8imagery and digital elevation model based on the Google Earth Engine[J ]. Earth System Science Data,2021 (5)), for cultivated lands of different morphologies, the crops planted in the cultivated lands are mixed, and the extraction of the above ways is still difficult to realize; secondly, the cultivation area of mountain areas is accompanied by forest lands, and the planting environment is relatively complex. To achieve fine extraction, it is necessary to resort to high resolution remote sensing images (see Zhang D, pan Y, zhang J, et al A generalized approach based on convolutional neural networks for large area cropland mapping at very high resolution [ J ]. Remote Sensing ofEnvironment,2020, 247:111912.). However, the high-resolution images are difficult to acquire and have low time resolution, and when the acquired images are exactly in autumn and winter, the spectral characteristics of the cultivated land on the images are similar to those of surrounding woodlands and the like, so that the cultivated land is difficult to identify.
With the increase of interpretation requirements, the accuracy requirement cannot be met by simply relying on image features, and the geoscience knowledge is increasingly applied to the remote sensing intelligent interpretation process as auxiliary data as direct data capable of reflecting the environment or distribution situation of the surface elements. The geographical division is an important means of geographical research, reflects the homogeneity inside the corresponding geographical area and the heterogeneity between areas by summarizing the geographical differentiation rules of the elements such as temperature, moisture, soil, biology and the like, is an important way to identify the geographical features and differences of the areas (see national academy of sciences, geographical institute of sciences, national institute of science [ M ], scientific press, 1959.) and has accumulated a lot of data since the 19 th century development. The application of geographical partitioning to remote sensing interpretation is an effective means of system incorporation into the knowledge of geography (see Zhang Bing, yang Xiaomei, gaojie, meng Yu, sun Xian, shore morning, ni Li. Geography cognition model and method for intelligent interpretation of remote sensing big data [ J ]. Mapping school report, 2022,51 (07): 1398-1415.). The current geochemistry partitioning method applied to remote sensing intelligent interpretation mainly comprises the following steps: 1) directly using the existing data of ecological partition, administrative partition, road network and the like, 2) judging the artificially generated partition scheme based on the experience of expert knowledge, and 3) dividing the threshold based on certain data (such as elevation data). The application of the above method in the interpretation process has the following disadvantages: 1) Most of the existing zone data are mostly aimed at large-scale drawing scenes and may not match with the research scale of the current interpretation target; 2) The dividing method according to expert knowledge depends on experience and has insufficient objectivity; 3) The single data can only reflect the space diversity rule of one geography element, and the threshold division is mostly represented as a rigid rule, so that fuzzy knowledge generated by mutual combination of multiple elements is ignored, and the actual surface distribution rule is possibly not met (see Liu Wei, wu Zhifeng, luo Jiancheng, sun Yingwei, wu Tianjun, zhou Nan, hu Xiaodong, wang Lingyu and Zhou Zhongfa. The deep learning supports a method [ J ] for partitioning and layering extraction of high-resolution remote sensing information of hilly and mountain areas, mapping school report, 2021,50 (01): 105-116). At present, although more data are accumulated in the geoscience partition, the geoscience partition method facing the remote sensing interpretation requirement is still not researched enough.
Disclosure of Invention
Aiming at the problems, in order to overcome the defects in the traditional extraction method, the invention provides a geologic partitioning method for extracting the cultivated land information, which comprehensively utilizes the idea of geologic partitioning, combines multi-source data, integrates geologic knowledge into a remote sensing interpretation process, delimits concentrated distribution areas of cultivated lands with different forms by a partitioning method, and further realizes high-precision automatic extraction of different cultivated land types based on the partitioning result.
The specific technical scheme of the invention is a geologic partitioning method for extracting tilling-oriented information, which is characterized by comprising the following steps:
1) Obtaining remote sensing image data and auxiliary data of an area for extracting farmland information, wherein the data comprise a high-resolution remote sensing image, a monthly average NDVI data set, building thematic data and DEM data, and establishing a farmland type interpretation mark and a classification system, wherein the classification system comprises first-class cultivated lands and non-cultivated lands, and the second-class cultivated lands further comprise regular cultivated lands, greenhouses, sloping cultivated lands and terraces;
2) Performing image multi-scale segmentation on the high-resolution remote sensing image of the region for extracting the cultivated land information, establishing a regular honeycomb grid in the region for extracting the cultivated land information, and collecting samples which are uniformly distributed in space in the grid to obtain samples of each cultivated land type and non-cultivated land;
3) First-level partitioning is carried out by utilizing month-average NDVI data and building thematic data, NDVI time sequence data is firstly obtained according to month-average NDVI data sets and is clustered to obtain clustered regions with concentrated distribution of cultivated lands and regions with concentrated distribution of non-cultivated lands, euclidean distance between each cultivated land sample and a building is calculated according to the building thematic data, a percentage threshold A is set, the total number of statistical samples is multiplied by the maximum value of Euclidean distance between the cultivated land samples of the number A and the building to obtain a distance threshold B, the regions within the Euclidean distance B of the building are considered to be the regions with concentrated distribution of the cultivated lands, the regions outside the Euclidean distance B of the building are considered to be the regions with concentrated distribution of the non-cultivated lands, the regions with concentrated distribution of the cultivated lands and the regions with concentrated distribution of the non-cultivated lands based on the building distance are obtained, the clustered region with concentrated arable land and the region with concentrated arable land based on building distance are spatially overlapped, the intersection of the obtained arable land regions is the region with concentrated arable land in the primary partition result, the other regions are the regions with concentrated arable land in the primary partition result, the next step is to divide the region with concentrated arable land into the concentrated distribution regions of initial regular arable land and greenhouse, terrace and hillside arable land according to the distribution characteristics of different arable land types in the terrain in combination with DEM data, further fill up the tiny holes in the initial region and remove the chip plaque, then superimpose the regions with concentrated distribution of arable land types, judge the arable land type attribute by adopting arable land samples in the intersection region, and finally obtain the secondary partition result, namely the actual regular arable land and greenhouse, concentrated distribution areas of terraces and hillsides;
4) Dividing a high-resolution remote sensing image of a region for extracting farmland information into 4 sub-regions which are respectively and intensively distributed with non-farmland, regular farmland, greenhouse, terraced fields and hillside farmland on the basis of a secondary-region dividing result, and respectively carrying out image multi-scale division in each sub-region;
5) Performing object-to-point-to-object sample conversion on the sample obtained in the step 2, specifically generating a point by using the sample center in the step 2, searching for the segmented object generated in the step 4) in which the point falls, and developing the segmented object into an object sample consistent with the point attribute;
6) And (3) respectively training a classification model in each subarea by utilizing the samples which are generated in the step (5) and fall into the subarea and a random forest algorithm, and realizing the extraction of farmland information under the support of the geoscience subarea.
Further, the specific process of the step 3) is as follows:
3.1 Constructing a cloud-free remote sensing image dataset of a long-time sequence, namely a monthly average NDVI dataset, calculating the NDVI based on the dataset, and carrying out median month synthesis on a calculation result to obtain NDVI time sequence data, wherein a calculation formula is shown as the following formula (I):
wherein, NIR is near infrared band of the current image, RED is RED band of the current image;
3.2 SG filtering the NDVI time series data in step 3.1 to realize noise removal and data smoothing;
3.3 K-Means clustering is carried out on the smoothed NDVI time sequence data, the number of the clusters is set to be 2, and the purpose is to realize the division of the concentrated distribution areas of the cultivated land and the non-cultivated land by a clustering Means so as to obtain the clustered concentrated distribution areas of the cultivated land and the non-cultivated land;
3.4 Calculating Euclidean distance between cultivated land samples and buildings according to the special data of the building area, wherein the percentage threshold A is equal to 95% and the distance threshold B is equal to 0.016 degrees, calculating the total number of samples multiplied by the maximum value of Euclidean distance between the cultivated land samples of the number A and the buildings to obtain the distance threshold B, considering the area within the Euclidean distance B of the buildings as the area intensively distributed in cultivated land, and the area outside the Euclidean distance B of the buildings as the area intensively distributed in non-cultivated land to obtain the area intensively distributed in cultivated land and the area intensively distributed in non-cultivated land based on the building distance;
3.5 Spatially superposing the clustered region with the concentrated distribution of the cultivated land and the region with the concentrated distribution of the non-cultivated land with the region with the concentrated distribution of the cultivated land after frequency statistics, wherein the intersection of the obtained region with the cultivated land is a first-level zoning result, namely the region with the concentrated distribution of the actual cultivated land and the region with the concentrated distribution of the non-cultivated land;
3.6 Calculating gradient and topography Relief based on DEM data so as to obtain elevation DEM, gradient Slope and topography Relief data in an area for extracting farmland information, then counting distribution ranges of different types of farmland samples of regular farmland, greenhouse, terraced fields and hillside farmland on elevation, gradient and topography Relief respectively, and taking intersection of 3 types of topography factors falling into the distribution ranges of 95% of the number of samples as a concentrated distribution area of each farmland type; the distribution area of the regular farmland is an area which simultaneously satisfies altitude below 600 meters, topography Relief less than 7 meters and gradient less than 7 degrees, and is expressed as { (DEM <600 m) & (Relief <7 m) & (Slope <7 °) }, the distribution area of the Slope farmland is { (DEM <900 m) & (Relief <14 m) & (Slope <21 °) }, the distribution area of the terrace is { (600 m ∈dem <900 m) & (Relief <21 °) }, and the distribution area of the greenhouse is { (DEM <600 m) & (Relief <7 °) }, and is consistent with the distribution area of the regular farmland, and the area in which the actual farmland is intensively distributed is further divided into an initial regular farmland, a greenhouse, a terrace and a concentrated distribution area of the Slope farmland;
3.7 Performing morphological open operation and close operation treatment on the area which is obtained in 3.6) and is used for intensively distributing the actual cultivated land and is further divided into an initial regular cultivated land and a concentrated distribution area of a greenhouse, a terrace and a slope cultivated land, filling tiny holes in the area and removing debris patches;
3.8 And 3.7, overlapping the areas which are intensively distributed in each cultivated land type, wherein the areas which are intensively distributed in each cultivated land type are overlapped, and because the boundary intersection generates some plagues which belong to two or more cultivated land types at the same time, judging the cultivated land type attribute of each cultivated land type by means of the sample, endowing the plagues with the corresponding cultivated land type attribute with the largest sample number which falls into the plagues, and combining the plagues with the same attribute, so as to finally obtain a secondary-stage division result, namely, the intensively distributed areas of the actual regular cultivated land, the greenhouse, the terrace and the sloping cultivated land.
The beneficial effects of the invention are as follows: 1) The invention integrates the classical geographic ideas of the geologic partition into the remote sensing interpretation process, combines multi-source data, provides a partition method for extracting farmland information, fully applies the geologic knowledge and the space relation of the ground object into the remote sensing interpretation process, and makes up the defect of purely relying on the remote sensing image spectral feature extraction method; 2) The method is characterized in that primary partition is firstly performed according to the spectral time sequence characteristics and the spatial distribution characteristics of the cultivated land and the non-cultivated land, secondary partition is further performed according to the topographic distribution characteristics of different cultivated lands, and a multi-level image partition method is provided by exploring the spatial diversity rules of different cultivated lands and is further used for guiding extraction of cultivated land information. By dividing the concentrated distribution areas of different types of cultivated lands, further training the classification models in each sub-area respectively, the interference of spectrum information of other ground object types can be weakened, and the problem of misclassification caused by identical foreign object different spectrums and identical bands of foreign objects is effectively solved; 3) The subarea division of the concentrated distribution of the cultivated land types of each form is realized through the geography subarea, and the optimal segmentation scale is further calculated for each subarea, so that the object boundary expression of the cultivated lands of different types is realized more accurately.
The method comprehensively utilizes multi-source data such as time sequence data, building thematic data, topographic data and the like, integrates the geologic partition thought and geologic knowledge into remote sensing intelligent interpretation, further divides the centralized distribution areas of different types of cultivated lands in a partition mode, largely avoids misclassification caused by homography or foreign matter homography, simultaneously realizes more accurate expression of the boundaries of all cultivated land types of objects, has obvious advantages, and has great potential in the aspect of automatically extracting cultivated land information in a large area with high precision.
Drawings
FIG. 1 is a flow chart of a method for geozoning for extracting tilling-oriented information in the invention;
FIG. 2 is a flowchart of image segmentation and sample collection in step 2 of the method for geozoning for extraction of tillable information according to the present invention;
FIG. 3 is a flowchart of the extraction of the tilled land information under support of the step 4-6 of the geozoning in the method of the invention;
FIG. 4 is a graph of multi-level geostatistical partitioning results fusing multi-source data;
FIG. 5 is a diagram of global and sub-region optimal segmentation scale preference;
FIG. 6 is a graph of results of extraction of tilling information under support of a geoscience partition;
FIG. 7 is a graph showing accuracy comparisons of classification results before and after geoscience partitioning.
Detailed Description
The technical scheme of the invention is further described below with reference to the attached drawings.
In a specific embodiment of the method, the area for extracting the cultivated land information is selected to be positioned at the junction of the Changping area, the Yanqing area and the Huaiyou area in Beijing city, and the topography fluctuation is larger than that of other areas in Beijing city. The remote sensing image data and the auxiliary data used comprise: the sentry No. 2 remote sensing image (cloud coverage < 10%) with resolution of 10 meters between 1 month and 1 month in 2020 and 31 months in 2020 is used for synthesizing an NDVI time sequence; the high-resolution No. 2 image with the resolution of 2 meters in 12 months in 2020 is used for extracting farmland information; building thematic data GISA (global impervious surface area) V2.0 with resolution of 30m in 2019; ALOS PALSAR DEM data at a resolution of 12.5 meters.
As shown in fig. 1, the detailed steps of the method for the geochemical partitioning for extracting the cultivated land information of the invention are as follows:
1) The method comprises the steps of knowing the spatial distribution condition of cultivated land types in an area for extracting cultivated land information, distributing four types of cultivated lands including regular cultivated lands, slope cultivated lands, terraced fields and greenhouses in the area for extracting the cultivated land information, carrying out non-cultivated lands such as forests, grasslands, urban areas, water bodies and the like, clearly determining remote sensing image features of each cultivated land type through field investigation, and establishing corresponding cultivated land type interpretation marks and classification systems. The classification system comprises a first class comprising cultivated land and non-cultivated land, and a second class comprising regular cultivated land, slope cultivated land, terraced fields and greenhouse.
2) And performing image multi-scale segmentation on the high-resolution remote sensing images for classification. And establishing a regular cellular grid in the area where the tilling information is extracted, and collecting samples which are uniformly distributed in space inside the grid. As shown in fig. 2.
2.1 Inputting the remote sensing image into the eCognition software, and dividing the image by using a Multiresolution Segmentation function;
2.2 Setting parameters 'Shape' and 'compact' as 0.2 and 0.6 respectively, setting the parameter 'Scale parameter' to be 100-300 in the area for extracting the tilling information, and setting the step length to be 10 for respectively carrying out segmentation, wherein the parameter 'Scale parameter' represents the size of a segmentation Scale;
2.3 Based on the segmentation results of the multiple segmentation scales, indicating the optimal parameters of the object segmentation effect by calculating the rate of change value ROC-LV (rates ofchange ofLV) of the Local Variance (LV) of the homogeneity of the image object under different segmentation scale parameters. When the rate of change value of LV is the maximum, namely the peak appears, the segmentation scale corresponding to the point is the optimal segmentation scale.
The calculation formula of ROC is:where L is the local variance value of the current segmentation scale and L-1 is the local variance value of the one-level lower segmentation scale. After calculation, the segmentation scale is selected to be 170;
2.4 Uniformly distributing hexagonal honeycomb grids with the same size in an area for extracting tilling information, wherein the hexagonal honeycomb grids are used for selecting training samples and verification samples, and the total number of the grids is 530;
2.5 For the selection of training samples, firstly, uniform sample selection is carried out in the grids, so that each grid is ensured to have one sample, and samples are added in the grids with complex ground feature distribution when necessary. A total of 650 training samples were generated, of which 366 non-cultivated lands, 155 regular cultivated lands, 71 sloping cultivated lands, 61 terraces, 27 greenhouses;
2.6 For the selection of verification samples, firstly, 3 samples are randomly generated in each grid, and if necessary, the samples are added in the grids with complex ground feature distribution in a random generation mode. In total, 1600 verification samples were generated, including 931 non-cultivated lands, 274 regular cultivated lands, 187 sloping cultivated lands, 151 terraces, and 57 greenhouses.
3) And combining a month average NDVI (Normalized Difference Vegetation Index) data set, building thematic data, DEM (Digital ElevationModel) and other multi-source data, and constructing a geoscience partition method by means of clustering, morphological processing, spatial data superposition analysis and the like, so as to realize multi-level geoscience partition facing interpretation requirements.
3.1 Based on a 10m resolution ratio of the second sentinel remote sensing image of year 2020, calculating the NDVI and synthesizing the calculated result in a median month, so as to construct 12 month NDVI time series data in an area for extracting farmland information, wherein the calculation formula of the NDVI is as follows:
wherein, NIR is near infrared band of the current image, RED is RED band of the current image;
3.2 SG filtering (Savitzky-Golay Filter) is performed on the NDVI time series data to realize noise removal and data smoothing;
3.3 Aiming at the difference of time series change of cultivated land and non-cultivated land, carrying out K-Means clustering on the smoothed NDVI time series data, wherein the clustering number is set to be 2. The K-Means clustering algorithm is implemented by PythonV3.7. The method aims at dividing the concentrated distribution areas of the two areas by using a clustering means to obtain clustered areas with concentrated distribution of cultivated lands and areas with concentrated distribution of non-cultivated lands;
3.4 Collecting the topical data of the building area according to the characteristic that most of cultivated land is distributed around the human activity area, and calculating the Euclidean distance between a cultivated land sample and a building. Counting the maximum value of the distance between the cultivated land samples accounting for 95% of the total cultivated land samples and the building, and displaying that the cultivated land samples accounting for 95% of the total cultivated land samples are distributed within 0.016 DEG from the building, so that a buffer zone is established at 0.016 DEG on the basis of the topical data of the building area, wherein the area within the buffer zone is the area in which the cultivated lands after frequency statistics are concentrated and distributed, and the areas outside the area in which the cultivated lands after frequency statistics are concentrated and distributed;
3.5 Spatially superposing the results of 3.3) and 3.4), wherein the intersection of the cultivated land areas of the two results is the final first-level zoning result (cultivated land concentrated distribution area and non-cultivated land concentrated distribution area);
3.6 By utilizing the characteristic of large difference of the topographic features of the spatial distribution areas of the cultivated lands in different forms, the initial concentrated distribution area of the regular cultivated lands, the greenhouse, the terraced fields and the hillside cultivated lands is obtained in the concentrated distribution area of the cultivated lands. First, gradient and topography Relief are calculated based on DEM data of 12.5m in resolution to obtain elevation (DEM), gradient (Slope) and topography Relief (Relief) data in an area where tilling information extraction is performed. Then, counting the distribution ranges of different types of cultivated land samples on the altitude, the gradient and the topography fluctuation respectively, and taking the intersection of 3 types of topography factors falling into the distribution range of 95% of the number of the samples as a concentrated distribution area of each cultivated land type; the distribution area of the regular farmland is an area which simultaneously satisfies altitude below 600 meters, topography Relief below 7 meters and gradient below 7 degrees, and is expressed as { (DEM <600 m) & (Relief <7 m) & (Slope <7 °) }, and the following is the same; the distribution area of the hillside fields is { (DEM <900 m) & (Relief <14 m) & (Slope <21 °) }, the distribution area of the terraces is { (600 m ∈DEM <900 m) & (Slope <21 °) }, the distribution area of the greenhouse is { (DEM <600 m) & (Relief <7 m) & (Slope <7 °) }, and the distribution area of the greenhouse is consistent with that of the regular hillside fields.
3.7 The morphological open and close processing of the region obtained in 3.6), the morphological processing being implemented in pythonv 3.7. Firstly, performing closed operation treatment, namely expanding and then corroding the area, so as to achieve the effect of eliminating small cavities and smoothing the outline of the object; wherein the convolution kernel of the closed operation is set to 10 x 10. On the basis of the closed operation result, performing open operation treatment, and firstly corroding and then expanding the area to achieve the effect of removing the clastic plaque; the convolution kernel in which the open operation is set to 15 x 15.
3.8 Overlapping the areas with concentrated distribution of each cultivated land type, judging the cultivated land type attribute by means of the samples at the moment because the boundary intersection generates some plagues belonging to two or more cultivated land types at the same time, endowing the plagues with the corresponding cultivated land type attribute with the largest number of the samples falling into the plagues, and merging the plagues with the same attribute to finally obtain a secondary grading result (the concentrated distribution area of the actual regular cultivated land and the greenhouse, terrace and slope cultivated land), as shown in figure 4.
4) And dividing the remote sensing image covering the area for extracting the tilling information into a plurality of subareas based on the secondary partition result, and carrying out self-adaptive optimization of the optimal scale of image segmentation in each subarea. The image segmentation and the selection method of the optimal scale are shown in 2.1-2.3 in the step 2). By calculation, the division scale in the final non-cultivated land partition is selected to be 200, the division scale in the regular cultivated land and greenhouse partition is selected to be 125, the division scale in the sloping cultivated land partition is selected to be 135, and the division scale in the terrace partition is selected to be 205, as shown in fig. 5.
5) And (3) performing object-point-object sample conversion on the samples obtained in the step (2) to realize sample alignment under different segmentation scales. Further checking whether the sample satisfies a uniform distribution among the types in each sub-area, and if not, performing sample optimization.
5.1 Generating a point at each sample center obtained in the step 2, converting the samples into sample specimens in batches, and keeping the attributes consistent;
5.2 Searching an object with a sample falling into the object based on the object obtained according to the optimal segmentation scale in each subarea, giving the object with the attribute of the sample, converting the sample batch into the object sample, and keeping the attribute consistent.
6) And respectively training a classification model in each subarea by using a sample falling into the subarea and a random forest algorithm, so as to realize tilling information extraction under the support of the geoscience subarea.
6.1 Random forest parameter setting, "n_detectors=100", other parameters default. And training the classification model in each subarea respectively, and finally obtaining the tilling information extraction result supported by the geoscience subarea, as shown in fig. 6.
6.2 And (3) comparing the classification results obtained by respectively training the models after the partitioning with the classification results of the non-partitioned areas in a mode of calculating the confusion matrix. The overall accuracy after partitioning was 90.81%, which is improved by 6% compared to the overall accuracy before partitioning. In addition, from the perspective of the precision of the producer and the precision of the user, the precision of the producer and the precision of the user after the partition are improved greatly compared with the precision of the producer and the precision of the user before the partition on three types of hillside farmlands, terraces and greenhouses, wherein the precision of the producer and the precision of the user after the partition are respectively improved by 6.88 percent and 28.34 percent, and the precision of the producer and the precision of the user after the partition are respectively improved by 3.53 percent and 16.56 percent, as shown in figure 7.
The method comprehensively utilizes multi-source data such as time sequence data, building thematic data, topographic data and the like, integrates the geologic partition thought and geologic knowledge into remote sensing intelligent interpretation, further divides the centralized distribution areas of different types of cultivated lands in a partition mode, largely avoids misclassification caused by homography or foreign matter homography, simultaneously realizes more accurate expression of the boundaries of all cultivated land types of objects, has obvious advantages, and has great potential in the aspect of automatically extracting cultivated land information in a large area with high precision.
While the invention has been disclosed in terms of preferred embodiments, the embodiments are not intended to limit the invention. Any equivalent changes or modifications can be made without departing from the spirit and scope of the present invention, and are intended to be within the scope of the present invention. The scope of the invention should therefore be determined by the following claims.
While the invention has been disclosed in terms of preferred embodiments, the embodiments are not intended to limit the invention. Any equivalent changes or modifications can be made without departing from the spirit and scope of the present invention, and are intended to be within the scope of the present invention. The scope of the invention should therefore be determined by the following claims.
Claims (2)
1. The method for the geochemical partitioning oriented to the extraction of the cultivated land information is characterized by comprising the following steps of:
1) Obtaining remote sensing image data and auxiliary data of an area for extracting farmland information, wherein the remote sensing image data and the auxiliary data comprise a high-resolution remote sensing image, a month-average NDVI data set, building thematic data and DEM data, and establishing a farmland type interpretation mark and a classification system, wherein the classification system comprises cultivated lands and non-cultivated lands in a first class, and the cultivated lands in a second class further comprises regular cultivated lands, greenhouse, sloping cultivated lands and terraces;
2) Performing image multi-scale segmentation on the high-resolution remote sensing image of the region for extracting the cultivated land information, establishing a regular honeycomb grid in the region for extracting the cultivated land information, and collecting samples which are uniformly distributed in space in the grid to obtain samples of each cultivated land type and non-cultivated land;
3) First-level partitioning is carried out by utilizing month-average NDVI data and building thematic data, NDVI time sequence data is firstly obtained according to month-average NDVI data sets and is clustered to obtain clustered regions with concentrated distribution of cultivated lands and regions with concentrated distribution of non-cultivated lands, euclidean distance between each cultivated land sample and a building is calculated according to the building thematic data, a percentage threshold A is set, the maximum value of Euclidean distance between cultivated land samples and the building, which is the total number of the samples multiplied by the number A, is counted to obtain a distance threshold B, the regions within the Euclidean distance B of the building are considered to be regions with concentrated distribution of cultivated lands, the regions outside the Euclidean distance B of the building are considered to be regions with concentrated distribution of non-cultivated lands, the regions with concentrated distribution of cultivated lands and the regions with concentrated distribution of non-cultivated lands based on the building distance are obtained, spatially superposing the clustered region with the region with concentrated arable land and the region with concentrated arable land based on the building distance, wherein the intersection of the obtained arable land regions is the region with concentrated arable land in the primary partition result, the other regions are the regions with concentrated arable land in the primary partition result, the regions with concentrated arable land in the secondary partition result are further divided into the regions with concentrated distribution of initial regular arable land, greenhouse, terrace and hillside arable land by combining DEM data according to the distribution characteristics of different arable land types, fine holes in the initial regions are further filled and debris patches are removed, the regions with concentrated distribution of each arable land type are superposed, the arable land type attribute of each arable land is judged by adopting each type arable land sample in the intersecting region, and finally a secondary partition result is obtained, namely, the actual regular farmland and the centralized distribution areas of the greenhouse, terraced fields and sloping fields;
4) Dividing a high-resolution remote sensing image of a region for extracting farmland information into 4 sub-regions which are respectively and intensively distributed with non-farmland, regular farmland, greenhouse, terraced fields and hillside farmland on the basis of a secondary-region dividing result, and respectively carrying out image multi-scale division in each sub-region;
5) Performing object-to-point-to-object sample conversion on the sample obtained in the step 2, specifically generating a point by using the sample center in the step 2, searching for the segmented object generated in the step 4) in which the point falls, and developing the segmented object into an object sample consistent with the point attribute;
6) And (3) respectively training a classification model in each subarea by utilizing the samples which are generated in the step (5) and fall into the subarea and a random forest algorithm, and realizing the extraction of farmland information under the support of the geoscience subarea.
2. The method for geochemical partitioning for extraction of information on cultivated land according to claim 1, wherein the specific process of step 3) is as follows:
3.1 Constructing a cloud-free remote sensing image dataset of a long-time sequence, namely a monthly average NDVI dataset, calculating the NDVI based on the dataset, and carrying out median month synthesis on a calculation result to obtain NDVI time sequence data, wherein a calculation formula is shown as the following formula (I):
wherein, NIR is near infrared band of the current image, RED is RED band of the current image;
3.2 SG filtering the NDVI time series data in step 3.1 to realize noise removal and data smoothing;
3.3 K-Means clustering is carried out on the smoothed NDVI time sequence data, the number of the clusters is set to be 2, and the purpose is to realize the division of the concentrated distribution areas of the cultivated land and the non-cultivated land by a clustering Means so as to obtain the clustered concentrated distribution areas of the cultivated land and the non-cultivated land;
3.4 Calculating Euclidean distance between cultivated land samples and buildings according to the special data of the building area, wherein the percentage threshold A is equal to 95% and the distance threshold B is equal to 0.016 degrees, calculating the total number of samples multiplied by the maximum value of Euclidean distance between the cultivated land samples of the number A and the buildings to obtain the distance threshold B, considering the area within the Euclidean distance B of the buildings as the area intensively distributed in cultivated land, and the area outside the Euclidean distance B of the buildings as the area intensively distributed in non-cultivated land to obtain the area intensively distributed in cultivated land and the area intensively distributed in non-cultivated land based on the building distance;
3.5 Spatially superposing the clustered region with the concentrated distribution of the cultivated land and the region with the concentrated distribution of the non-cultivated land with the region with the concentrated distribution of the cultivated land after frequency statistics, wherein the intersection of the obtained region with the cultivated land is a first-level zoning result, namely the region with the concentrated distribution of the actual cultivated land and the region with the concentrated distribution of the non-cultivated land;
3.6 Calculating gradient and topography Relief based on DEM data so as to obtain elevation DEM, gradient Slope and topography Relief data in an area for extracting farmland information, then counting distribution ranges of different types of farmland samples of regular farmland, greenhouse, terraced fields and hillside farmland on elevation, gradient and topography Relief respectively, and taking intersection of 3 types of topography factors falling into the distribution ranges of 95% of the number of samples as a concentrated distribution area of each farmland type; the distribution area of the regular farmland is an area which simultaneously satisfies altitude below 600 meters, topography Relief less than 7 meters and gradient less than 7 degrees, and is expressed as { (DEM <600 m) & (Relief <7 m) & (Slope <7 °) }, the distribution area of the Slope farmland is { (DEM <900 m) & (Relief <14 m) & (Slope <21 °) }, the distribution area of the terrace is { (600 m ∈dem <900 m) & (Relief <21 °) }, and the distribution area of the greenhouse is { (DEM <600 m) & (Relief <7 °) }, and is consistent with the distribution area of the regular farmland, and the area in which the actual farmland is intensively distributed is further divided into an initial regular farmland, a greenhouse, a terrace and a concentrated distribution area of the Slope farmland;
3.7 Performing morphological open operation and close operation treatment on the area which is obtained in 3.6) and is used for intensively distributing the actual cultivated land and is further divided into an initial regular cultivated land and a concentrated distribution area of a greenhouse, a terrace and a slope cultivated land, filling tiny holes in the area and removing debris patches;
3.8 And 3.7, overlapping the areas which are intensively distributed in each cultivated land type, wherein the areas which are intensively distributed in each cultivated land type are overlapped, and because the boundary intersection generates some plagues which belong to two or more cultivated land types at the same time, judging the cultivated land type attribute of each cultivated land type by means of the sample, endowing the plagues with the corresponding cultivated land type attribute with the largest sample number which falls into the plagues, and combining the plagues with the same attribute, so as to finally obtain a secondary-stage division result, namely, the intensively distributed areas of the actual regular cultivated land, the greenhouse, the terrace and the sloping cultivated land.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211627162.9A CN116206197B (en) | 2022-12-16 | 2022-12-16 | Ground science partitioning method for farmland information extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211627162.9A CN116206197B (en) | 2022-12-16 | 2022-12-16 | Ground science partitioning method for farmland information extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116206197A CN116206197A (en) | 2023-06-02 |
| CN116206197B true CN116206197B (en) | 2024-02-20 |
Family
ID=86510307
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211627162.9A Active CN116206197B (en) | 2022-12-16 | 2022-12-16 | Ground science partitioning method for farmland information extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116206197B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112164062A (en) * | 2020-10-29 | 2021-01-01 | 河海大学 | Wasteland information extraction method and device based on remote sensing time sequence analysis |
| CN114663771A (en) * | 2022-04-14 | 2022-06-24 | 中国农业科学院农业资源与农业区划研究所 | Mountain farmland intelligent extraction method based on partition layering theory |
| CN115468917A (en) * | 2022-08-11 | 2022-12-13 | 中国农业大学 | Method and system for extracting crop information of farmland plot based on high-resolution remote sensing |
-
2022
- 2022-12-16 CN CN202211627162.9A patent/CN116206197B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112164062A (en) * | 2020-10-29 | 2021-01-01 | 河海大学 | Wasteland information extraction method and device based on remote sensing time sequence analysis |
| CN114663771A (en) * | 2022-04-14 | 2022-06-24 | 中国农业科学院农业资源与农业区划研究所 | Mountain farmland intelligent extraction method based on partition layering theory |
| CN115468917A (en) * | 2022-08-11 | 2022-12-13 | 中国农业大学 | Method and system for extracting crop information of farmland plot based on high-resolution remote sensing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116206197A (en) | 2023-06-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pan et al. | Mapping cropping intensity in Huaihe basin using phenology algorithm, all Sentinel-2 and Landsat images in Google Earth Engine | |
| Keshtkar et al. | Land-cover classification and analysis of change using machine-learning classifiers and multi-temporal remote sensing imagery | |
| Walsh et al. | A multiscale analysis of LULC and NDVI variation in Nang Rong district, northeast Thailand | |
| Birhanu et al. | Land use/land cover change along elevation and slope gradient in highlands of Ethiopia | |
| CN107330413B (en) | A kind of mother drug plants recognition methods based on remote sensing technology | |
| Rogan et al. | Integrating GIS and remotely sensed data for mapping forest disturbance and change | |
| CN114387516B (en) | Single-season rice SAR (synthetic aperture radar) identification method for small and medium-sized fields in complex terrain environment | |
| Lisita et al. | Identifying potential areas of Cannabis sativa plantations using object-based image analysis of SPOT-5 satellite data | |
| Xie et al. | Land use and land cover change within the Koshi River Basin of the central Himalayas since 1990 | |
| Wang et al. | Research on cropping intensity mapping of the Huai River Basin (China) based on multi-source remote sensing data fusion | |
| Xing et al. | Mapping irrigated, rainfed and paddy croplands from time-series Sentinel-2 images by integrating pixel-based classification and image segmentation on Google Earth Engine | |
| Rimal et al. | Crop cycles and crop land classification in Nepal using MODIS NDVI | |
| CN113570273A (en) | Spatial method and system for irrigation farmland statistical data | |
| Dong et al. | Fine mapping of key soil nutrient content using high resolution remote sensing image to support precision agriculture in Northwest China | |
| CN116206197B (en) | Ground science partitioning method for farmland information extraction | |
| CN115830464A (en) | Plateau mountain agricultural greenhouse automatic extraction method based on multi-source data | |
| Jande et al. | Assessment of land use and land cover changes and urban expansion using remote sensing and GIS in Gboko, Benue State, Nigeria | |
| Pan et al. | Urban expansion and intra-urban land evolution as well as their natural environmental constraints in arid/semiarid regions of China from 2000–2018 | |
| Lee et al. | Evaluation of crop mapping on fragmented and complex slope farmlands through random forest and object-oriented analysis using unmanned aerial vehicles | |
| Xu | RETRACTED ARTICLE: Application of remote sensing image classification based on adaptive Gaussian mixture model in analysis of mountain environment features | |
| Wu et al. | Simulations of spatial patterns and species distributions in sandy land using unmanned aerial vehicle images | |
| Nguyen et al. | Land Use/Land Cover Changes from 1995 to 2017 in Trang Bang, Southern Vietnam | |
| Compaoré | The impact of savannah vegetation on the spatial and temporal variation of the actual evapotranspiration in the Volta Basin, Navrongo, Upper East Ghana | |
| Chandna et al. | Assessment of cropping intensity dynamics in Odisha using multitemporal Landsat TM and OLI images | |
| Bao et al. | A fine digital soil mapping by integrating remote sensing-based process model and deep learning method in Northeast China |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |