CN115187866A - Ecological pattern spot extraction method based on high-resolution image - Google Patents
Ecological pattern spot extraction method based on high-resolution image Download PDFInfo
- Publication number
- CN115187866A CN115187866A CN202210880187.3A CN202210880187A CN115187866A CN 115187866 A CN115187866 A CN 115187866A CN 202210880187 A CN202210880187 A CN 202210880187A CN 115187866 A CN115187866 A CN 115187866A
- Authority
- CN
- China
- Prior art keywords
- image
- vector
- extraction
- ecological
- samples
- 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.)
- Pending
Links
Images
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- 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
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an ecological pattern spot extraction method based on a high-resolution image, which comprises the following steps: acquiring a high-resolution image, and extracting road network, water system and terrain line data; taking the road network, the water system and the topographic data as extraction masks, and splitting an image into a plurality of independent subtask areas for regional ecological pattern spot extraction; according to the image characteristics in each subtask area, performing layered extraction on the earth surface elements; and updating the extracted ecological pattern spots to the same vector file according to the sequence from large scale to small scale of the vector result of each land type extracted in a layering way to form an ecological pattern spot vector of all elements, and obtaining a final ecological system pattern spot vector map. According to the ecological map spot extraction method, the complex earth surface can be simplified, and the elements with similar characteristics are concentrated in a plurality of relatively independent sub-regions, so that the difficulty in extracting the ground features is reduced.
Description
Technical Field
The invention relates to the technical field of ecology and remote sensing, in particular to a 'image spot level' ecosystem extraction method based on a high-resolution remote sensing image.
Background
The ecosystem is a unified whole formed by organisms and the environment in a certain space of the nature. The ecosystem can range from large to small, and the representation of the ecosystem required at different scales is also inconsistent. At present, in the national scale, provincial scale and urban scale, the earth surface is mostly divided into different ecosystem types by a pixel-oriented classification mode with a medium-resolution image as a data base. The method can quickly acquire the type of the ecosystem under the condition of a limited scale and has higher accuracy. However, on a county scale, due to the reduction of the representation range, the image representation information with medium resolution is also greatly reduced, and the content represented in a single pixel is complex and various, so that the ecosystem type is not pure. Compared with the medium-low resolution images, the high-resolution remote sensing images have richer spatial information characteristics in space and have more expression contents than the medium-low resolution images. If the classification method facing to the pixel is still adopted for the type division of the ecosystem, the situation that the pattern spots are broken and discontinuous can be caused, and the appearance is not attractive visually. The high-resolution images can clearly distinguish each independent space geographic entity, and the situation of broken pattern spots can be effectively reduced by taking the specific geographic entity as an object for extraction. Meanwhile, because each image spot represents a unique object, the shape and the boundary of the object are highly consistent with the representation content on the image, compared with a regular grid, the internal information of the object is unique, and the situation that a plurality of ecosystem types are mixed in one image spot does not exist. At present, a plurality of methods for extracting the earth surface type by using the high-resolution remote sensing image exist, and good results are obtained from the traditional object-oriented classification method and the pixel-level classification method. Most of the extraction methods only extract a certain type of the earth surface, such as the extraction of buildings and the extraction of cultivated land plots. When multiple elements need to be extracted simultaneously, the effect of extraction by one method is often unsatisfactory, and a large number of situations of wrong extraction and extraction omission occur.
How to accurately extract the type of the ecosystem through the high-resolution image becomes difficult. The earth is used as a complex giant system, the earth surface types are rich and various, and the extraction of the earth surface elements under complex conditions by means of a method is unrealistic. How to decompose the complex earth surface and improve the accuracy of the extraction result is a difficult problem to be solved in the field.
Disclosure of Invention
The invention aims to provide an ecological map spot extraction method based on a high-resolution image, which can simplify a complex earth surface, concentrate elements with similar characteristics in a plurality of relatively independent sub-regions and reduce the difficulty of extracting ground objects.
In order to achieve the above object, an embodiment of the present invention provides an ecological spot extraction method based on high resolution images, which includes the following steps: acquiring a high-resolution image, and extracting road network, water system and terrain line data; taking the road network, the water system and the topographic data as extraction masks, splitting the image into a plurality of independent subtask areas for regional ecological pattern spot extraction; according to the image characteristics in each subtask area, performing layered extraction on surface elements, wherein the surface elements comprise cultivated land, buildings, water surfaces, grasslands, forest lands and bushes; and updating the extracted ecological pattern spots to the same vector file according to the sequence from large scale to small scale of the vector result of each land type extracted in a layering way to form an ecological pattern spot vector of all elements, and obtaining a final ecological system pattern spot vector map.
In other words, the main concept of the invention is to deconstruct the complex earth surface according to the concept of geographic layering and zoning, divide the image into a plurality of independent subtask areas by using road network, water system and terrain data, stratify the image according to the element characteristics in each subtask area, buildings, water areas, cultivated lands, forests and the like, select corresponding methods to extract the image in sequence according to different visual characteristics of each layer of elements, and finally merge the data of each layer to obtain the ecological map spot vector data of the complete area.
In one or more embodiments of the present invention, acquiring a high resolution image and extracting road network, water system and topographic line data includes: acquiring a high-resolution remote sensing image of a target area; carrying out image preprocessing on the original data of the high-resolution remote sensing image, wherein the image preprocessing comprises orthorectification, radiometric calibration, atmospheric correction and cloud removal; drawing a main road network river system according to the surface features in the image; and calculating ridge lines and valley lines of the image area, and forming a terrain net.
In one or more embodiments of the present invention, the splitting an image into a plurality of independent subtask areas using the road network, the water system, and the terrain data as extraction masks includes: updating the road and the terrain network into a unified layer file to form a complete subarea control network; and utilizing a partition control network to cut the image to form a plurality of subtask areas with small data size.
In one or more embodiments of the present invention, the performing hierarchical extraction on the surface elements according to the image features in each subtask area includes: sample preparation, model training and image prediction, grid vectorization and vector post-processing, and vegetation extraction.
In the sample preparation, selecting sample points of regular farmland, terraced fields, sloping farmland, building areas and water surface according to the image characteristics and the types of the farmland in each subtask area; according to different types of sample points, cutting the sample points into images with fixed sizes and corresponding vector frames; performing feature drawing on the cut sample file, and giving non-zero attribute information to the drawn vector; and converting the samples with the attributes into a binary image for training.
In model training and image prediction, inputting samples into corresponding convolutional neural networks for training according to the types of the samples to obtain corresponding farmland, building and water surface extraction models; and inputting the images and the extraction models of the sub-regions into the corresponding neural networks, predicting the images and obtaining the predicted intensity maps of the cultivated land, the building and the water surface.
Performing vectorization operation on the predicted intensity map in grid vectorization and vector post-processing, wherein the edge intensity map extracts skeleton lines by using related tools, performs vectorization after line disconnection is connected, and converts the vector lines into a vector plane after vector line simplification and straightening to obtain a map spot; the surface predicted intensity graph is converted into a vector surface result after expansion, hole filling and binarization processing. In vegetation extraction, forest, grassland and shrub pattern spots are obtained, and a vector result of vegetation extraction is obtained.
In one or more embodiments of the invention, the vegetation extraction comprises: taking cultivated land, buildings and water surface as masks and terrain lines as constraint conditions, and carrying out multi-scale segmentation on the remaining image range; on the basis of the segmentation result, selecting forests, shrubs and grasslands as classification samples, and distinguishing all segmentation pattern spots by adopting a random forest method to obtain the forests, the grasslands and the shrubs; and carrying out edge simplification and smoothing on the extraction result to obtain a final vegetation extraction vector result.
In one or more embodiments of the present invention, forming an ecological map spot vector of a full element for the vector result of each land type extracted in a hierarchical manner includes: merging the same type of pattern spots of each subtask area to obtain an extraction result vector of each ecosystem type of the complete area; updating the extracted ecological pattern spots of forest-bush-grassland-water surface-building-cultivated land on the same vector file according to an updating sequence from large scale to small scale, and then updating roads and water systems in the vector file to form an ecological pattern spot vector of all elements; and carrying out single-component to multi-component operation on the updated image layer to eliminate fine broken spots and long and narrow pattern spots in the updated image layer so as to obtain a final ecosystem image.
In one or more embodiments of the present invention, the inputting samples into corresponding convolutional neural networks for training according to the types of the samples to obtain corresponding farmland, building and water surface extraction models includes: and inputting a farmland sample into the RCF network for training, inputting a building sample into the D-LinkNet network for training, and inputting a water surface sample into the U-Net network for training to obtain corresponding farmland, building and water surface extraction models.
In one or more embodiments of the present invention, the arable land sample is input into the RCF network for training, including: performing overall analysis on the global farmland plots, uniformly selecting farmland samples in an image range, drawing farmland boundary information in the farmland samples, and converting the farmland boundary information into binary results serving as label data serving as training data of an edge monitoring network; through training boundary characteristics, obtaining an extraction model of farmland boundary information, predicting the cut images of each task area, and obtaining a boundary prediction strength result graph of each task area; obtaining a binary image of a farmland prediction result through skeleton line extraction and broken line connection; and obtaining a cultivated land extraction vector result through post-processing such as grid vectorization, vector straightening, smoothing and the like.
In one or more embodiments of the invention, the building sample is input into a D-LinkNet network for training, including: analyzing the residence places in each subtask area, taking the residence places with similar characteristics as the same type of extraction object, making building samples with different image characteristics, drawing building areas in the samples and distinguishing the building areas from the background by using the attribute values; converting the drawn building area sample into binary label data according to the attribute value; inputting the building binary label data into a D-LinkNet network for training to obtain an extraction model of a residential area; predicting the image of the subarea by using the network to obtain a residence prediction intensity map of the corresponding subarea; and obtaining a final vector extraction result of the building area through grid vectorization and vector post-processing operation.
In one or more embodiments of the invention, the input of the water surface sample into the U-Net network for training comprises: analyzing the characteristics of the river water surface, the lake water surface and the pool water surface in the image, selecting representative rivers, lakes and pools as samples, drawing the water surfaces in the samples, and marking the difference from the background according to the attributes; converting the drawn sample into a binary icon label file according to the attributes, inputting the binary icon label file into a U-Net network, and training the characteristics to obtain a water surface extraction model; predicting the image of the subarea by using the water surface extraction model to obtain a water surface prediction intensity graph of the image; and obtaining a final water surface vector extraction result by adopting a grid vectorization and vector post-processing tool.
Compared with the prior art, the method for extracting the ecological image spots based on the high-resolution images has the following advantages that:
(1) According to the ecological image spot extraction method based on the high-resolution image, the data are partitioned by adopting the partitioning idea, the complex earth surface can be simplified, elements with similar characteristics are concentrated in a plurality of relatively independent sub-areas, and the difficulty in ground feature extraction is reduced. Meanwhile, the data volume of general high-resolution images is large, when the extraction area is large, direct extraction usually wastes time and labor, even the state that extraction cannot be performed occurs, and the requirement on hardware is high. After the partition processing is carried out, the image with larger data volume is divided into a plurality of small areas, the data volume of a single image can be effectively dispersed, the possibility of problems caused by large data volume in the extraction process is reduced, meanwhile, multi-machine multi-task parallel processing can be carried out, and the extraction time is obviously shortened.
(2) By utilizing the layering thought, the earth surface is divided into types with respective characteristics, and compared with other methods which do not carry out partition layering and direct extraction, the targeted extraction method selected according to the specific visual characteristics of each type has obvious improvement on the effects of wrong extraction and missed extraction.
(3) The addition of the vector post-processing enables the finally obtained result to have more beautiful shape; and after the layer is updated, complete data support is provided for the production and subsequent application of the thematic map.
Drawings
Fig. 1 is a simplified flowchart of an ecological patch extraction method based on high resolution images according to an embodiment of the present invention;
FIG. 2 is a detailed flowchart of an ecological patch extraction method based on high resolution images according to an embodiment of the present invention;
FIG. 3 is a sample superimposed grass ecosystem vector pattern patch according to an embodiment of the invention;
FIG. 4 is a sample superimposed field ecosystem vector pattern patch according to an embodiment of the present invention.
Detailed Description
The following detailed description of the present invention is provided in conjunction with the accompanying drawings, but it should be understood that the scope of the present invention is not limited to the specific embodiments.
Throughout the specification and claims, unless explicitly stated otherwise, the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated element or component but not the exclusion of any other element or component.
As shown in fig. 1 to 2, the method for extracting ecological patches based on high resolution images according to the preferred embodiment of the present invention includes the following steps S1 to S4.
In step S1, a high-resolution image is acquired, and road network, water system, and topographic line data are extracted.
Specifically, the trunk road and water system data in the image area range can be acquired from the public data, and the GIS piece is used for supplementing and drawing the missing road and water system by combining the actual image characteristics to form the water system and road network data of the image area. And extracting a terrain line in an image range by using a Digital Elevation Model (DEM), and synthesizing control network data in the image range according to a road network, a water system and the terrain data for subsequent image splitting.
In one embodiment, step S1 may include: 1) Acquiring high-resolution remote sensing images of a target area, such as Google images, GF2, quickbird and the like; 2) Carrying out image preprocessing such as orthorectification, radiometric calibration, atmospheric correction and cloud removal on the original data; 3) Drawing a main road network river system according to the surface features in the image; 4) Calculating ridge lines and valley lines of the image area by using the DEM, and forming a terrain network; 5) Updating the road and the terrain network into a unified layer file to form a complete subarea control network; 6) And cutting the image by using a partition control network to form a plurality of subtask areas with small data size.
In step S2, the road network, the water system, and the topographic data are used as extraction masks, and the image is divided into a plurality of independent subtask areas (a plurality of blocks) for extracting the ecological patches in the sub-areas.
Specifically, in step S2, the road and the terrain network may be updated to the unified layer file to form a complete partition control network; and the image is cut by using a partition control network to form a plurality of subtask areas with small data size.
In step S3, according to the image features in each subtask area, surface elements including cultivated land, buildings, water surface, grassland, forest land, and shrubs are extracted hierarchically.
Specifically, in the ecological map spot extraction method of the present invention, the surface elements are divided into a plurality of different types according to the difficulty of extraction, and the extraction is performed sequentially. The layered extraction mainly comprises the following steps:
(1) Sample preparation: selecting sample points of regular farmland, terraced fields, slope farmland, building areas and water surface according to the image characteristics in each sub-area and the type of the farmland; the samples need to be evenly distributed and in moderate numbers. And according to different types of sample points, cutting the sample points into images with fixed sizes and corresponding vector frames. Wherein, the regular farmland and the terraced fields are line vector files, and the slope farmland, the buildings and the water surface are surface vector files. And (3) performing feature drawing on the cut sample file, drawing edge feature information in a land block by using a rule farmland and a terrace, drawing the range of internal texture feature information by using a slope farmland, a building and a water surface, and endowing non-zero attribute information to the drawn vector. And finally converting the samples endowed with the attributes into a binary image for training.
(2) Model training and image prediction: and inputting the samples into a corresponding convolutional neural network for training according to the types of the samples. The farmland samples are input into the RCF network for training, the building samples are input into the D-LinkNet network for training, and the water surface samples are input into the U-Net network for training to obtain corresponding farmland, building and water surface extraction models. And inputting the images and the extraction models of the sub-regions into the corresponding neural networks, and predicting the images to obtain a prediction intensity map of the cultivated land, the building and the water surface.
(3) Grid vectorization and vector post-processing: and performing vectorization operation on the predicted intensity map. Extracting skeleton lines from the edge intensity graph by using related tools, connecting broken lines, carrying out vectorization, simplifying and straightening vector lines, and converting the vector lines into a vector plane to obtain a graph spot; the surface predicted intensity map is converted into a vector surface result after expansion, hole filling and binarization processing.
(4) Vegetation extraction: and (4) taking cultivated land, buildings and water surface as masks and terrain lines as constraint conditions, and carrying out multi-scale segmentation on the residual image range. On the basis of the segmentation result, a visual discrimination mode is adopted to select the forest, the bush and the grassland as classification samples, and a random forest method is adopted to discriminate all segmentation pattern spots to obtain the forest, the grassland and the bush pattern spots. And performing edge simplification and smoothing on the extraction result to obtain a final vegetation extraction vector result.
The detailed description is given below by sequentially extracting cultivated land, buildings, water surface, grassland, forest land and shrubs from the image of the sub-area according to the scale from small to large.
Cultivated land parcel extraction
The cultivated land extracted here takes the planting land block with obvious cultivated land boundary characteristics as the object, and the cultivated land block can be clearly distinguished in different image time phases. The RCF edge monitoring network has strong advantages in extracting fine boundary information, has more and abundant boundary characteristics in cultivated land blocks, and can be extracted by using the edge monitoring network better. The extraction steps follow the basic steps of deep learning: sample preparation, model training, image prediction, grid vectorization and vector post-processing.
Firstly, carrying out overall analysis on global farmland plots, uniformly selecting farmland samples in an image range, drawing farmland boundary information in the farmland samples by using GIS software, and converting the farmland boundary information into binary results serving as label data serving as training data of an edge monitoring network. The method comprises the steps of obtaining an extraction model of farmland boundary information through training boundary characteristics, predicting images of all cut task areas to obtain a boundary prediction strength result graph of all the task areas, obtaining a binary graph of a farmland prediction result through skeleton line extraction and broken line connection on the basis, and finally obtaining a farmland extraction vector result through post-processing such as grid vectorization, vector straightening, smoothing and the like.
Construction of buildings
Buildings are mainly distributed in places with gathered population, and most buildings appear in pieces. The buildings in the city are extracted according to the specific definition of the town ecosystem in the norm. Although a building has clear shape characteristics, the roof and the ground are similar in material in an image, so that the shape of a part of the building is easily damaged, and the specific shape of the building cannot be clear. The similar materials cause that the internal features in the building areas are consistent, so the internal texture features are adopted as extraction conditions, the buildings have different shapes and colors, and the difference between the ground among the buildings and the features of the buildings is not large, so the internal texture is adopted as the features for extraction, and the extraction process follows the basic steps of sample drawing, label making, sample training, image prediction and vector extraction.
The method comprises the steps of firstly analyzing the residence places in each subtask area, taking the residence places with similar characteristics as the same type of extraction objects, making building samples with different image characteristics, drawing building areas in the samples and distinguishing the building areas from the background by using attribute values. And converting the drawn building area sample into binary label data according to the attribute value. The D-LinkNet obtains a satisfactory effect on road extraction, and according to the test, under the condition of artificial earth surface and complex internal texture characteristics, the position and form information of the target to be extracted can be accurately acquired. Inputting the building binary label data into a D-LinkNet network for training to obtain an extraction model of a residential area; predicting the image of the subarea by using the network to obtain a residence prediction intensity map of the corresponding subarea; and finally, obtaining a final vector extraction result of the building area through grid vectorization and vector post-processing operation.
Surface of water
On the high-resolution remote sensing image, the river water surface, the lake water surface, the pool water surface and the non-water surface earth surface around the river water surface, the lake water surface and the pool water surface have obvious transition characteristics, and the boundaries of the water surfaces can be visually judged through naked eyes. But the earth surface of the non-water surface part is different, and the boundary characteristics can be obviously different. The features of different water surface interiors are similar and homogeneous, and compared with boundaries, the method is simpler, so that the texture of the water surface interior is adopted as an extraction condition. The extraction steps follow the steps of sample drawing, label making, label training, image prediction and vector extraction.
Firstly, analyzing the characteristics of the river water surface, the lake water surface and the pond water surface in the image, selecting representative rivers, lakes and ponds as samples, drawing the water surfaces in the samples, and marking the differences from the background according to the attributes. The U-Net network is a classical image segmentation network, and can quickly acquire position and form information of an internal homogeneous water surface. And converting the drawn sample into a binary icon label file according to the attributes, inputting the binary icon label file into a U-Net network, and training the characteristics to obtain a water surface extraction model. And predicting the image of the subarea by using the model to obtain a water surface prediction intensity map of the image. And obtaining a final water surface vector extraction result by adopting a grid vectorization and vector post-processing tool.
Shrub and forest grass
And after the extraction of the surface types is finished, taking the extraction result as a mask, and performing multi-scale segmentation on the rest area in the image to obtain a primary brushwork, grassland and wetland ecosystem pattern spot. On the basis of the pattern spots, a decision tree classification system is constructed according to the difference of the characteristics of forests, bushes and grasslands on vegetation, soil and the like, and the type of an ecological system for segmenting the pattern spots is further determined.
Woodland and grassland are used as natural earth surfaces, and have significant difference with other semi-natural and artificial earth surfaces. After the farmland, the buildings and the water surface are extracted, the elements are used as masks, the unextracted area outside the masks is subjected to multi-scale segmentation, the natural vegetation is segmented into smaller-scale patches, and the segmented patches are classified by a random forest classification method by combining a normalized vegetation index (NDVI), a vegetation coverage (FVI) and soil erosion parameters, so that the classification results of forests, bushes and grasslands are obtained.
According to the difference of spectral features of forests, shrubs and grasslands on the remote sensing images, corresponding to the three vegetation types, selecting typical representative regions as classification samples respectively, constructing a random forest classifier by using DEM and vegetation indexes as auxiliary factors, classifying the segmented patches to obtain the confidence coefficient of the classification result, correcting the attribute of the patches by contrasting the features on the images for the patches with low confidence coefficient, adding the patches into the classification samples for iterative classification, and continuously optimizing the classification results of the shrubs, the forests and the grasslands.
In step S4, the vector results of each land type extracted in a layered mode are sequentially updated to forest and grass pattern spots by using related tools according to the sequence from large scale to small scale, finally roads and water systems serving as control data are updated to the ecological system pattern spot results, and the fine broken spots and the long and narrow pattern spots existing in the ecological system pattern spots are processed to obtain a final ecological system pattern spot vector map.
Specifically, the same type of patches of each subtask area are merged to obtain extraction result vectors of each ecosystem type of the complete area. Updating the extracted ecological pattern spots to the same vector file according to an updating sequence (forest-bush-grassland-water surface-building-cultivated land) from large scale to small scale, and then updating roads and water systems to the file to form the ecological pattern spot vector of the whole element. And carrying out single-component to multi-component operation on the updated image layer to eliminate fine broken spots and long and narrow pattern spots in the updated image layer so as to obtain a final ecosystem image.
Fig. 3 and 4 show examples of superimposed grass ecosystem vector patches and superimposed field ecosystem vector patches according to an embodiment of the present invention.
According to the ecological image spot extraction method based on the high-resolution image, the data are partitioned by adopting the partitioning idea, the complex earth surface can be simplified, the elements with similar characteristics are concentrated in a plurality of relatively independent sub-areas, and the difficulty in extracting the ground features is reduced.
The foregoing description of specific exemplary embodiments of the invention has been presented for the purposes of illustration and description. It is not intended to limit the invention to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. The exemplary embodiments were chosen and described in order to explain certain principles of the invention and its practical application to enable one skilled in the art to make and use various exemplary embodiments of the invention and various alternatives and modifications. It is intended that the scope of the invention be defined by the claims and their equivalents.
Claims (10)
1. An ecological pattern spot extraction method based on a high-resolution image is characterized by comprising the following steps:
s1: acquiring a high-resolution image, and extracting road network, water system and terrain line data;
s2: taking the road network, the water system and the topographic data as extraction masks, splitting the image into a plurality of independent subtask areas for regional ecological pattern spot extraction;
s3: according to the image characteristics in each subtask area, performing layered extraction on surface elements, wherein the surface elements comprise cultivated land, buildings, water surfaces, grasslands, forest lands and bushes; and
s4: and updating the extracted ecological pattern spots to the same vector file according to the sequence from large scale to small scale for the vector result of each land type extracted by layers to form an ecological pattern spot vector of all elements, and obtaining a final ecological system pattern spot vector map.
2. The method for extracting ecological patches based on high-resolution images as claimed in claim 1, wherein the step S1 comprises:
acquiring a high-resolution remote sensing image of a target area;
carrying out image preprocessing on the original data of the high-resolution remote sensing image, wherein the image preprocessing comprises orthorectification, radiometric calibration, atmospheric correction and cloud removal;
drawing a main road network river system according to the surface features in the image; and
and calculating ridge lines and valley lines of the image area and forming a terrain net.
3. The method for extracting ecological patches based on high-resolution images as claimed in claim 1, wherein the step S2 comprises:
updating the road and the terrain network into a unified layer file to form a complete subarea control network; and
and cutting the image by using a partition control network to form a plurality of subtask areas with small data size.
4. The ecological patch extracting method based on high resolution image as claimed in claim 1, wherein the step S3 comprises:
sample preparation: selecting sample points of regular farmland, terraced fields, sloping farmland, building areas and water surface according to the image characteristics and the types of the farmland in each subtask area; cutting the sample points into images with fixed sizes and corresponding vector frames according to different types of sample points; performing feature drawing on the cut sample file, and endowing non-zero attribute information to a drawn vector; converting the samples with the attributes into a binary image for training;
model training and image prediction: inputting the samples into corresponding convolutional neural networks for training according to the types of the samples to obtain corresponding farmland, building and water surface extraction models; inputting the images and the extraction models of the sub-regions into corresponding neural networks, predicting the images and obtaining predicted intensity maps of cultivated land, buildings and water surfaces;
grid vectorization and vector post-processing: vectorizing the predicted intensity map, wherein the edge intensity map utilizes related tools to extract skeleton lines, vectorizing is carried out after line breakage is connected, vector lines are simplified and straightened, and then the vector lines are converted into vector planes, and a map spot is obtained; the surface prediction intensity graph is converted into a vector surface result after expansion, hole filling and binarization processing are carried out; and
vegetation extraction: and acquiring forest, grassland and bush pattern spots, and acquiring a vector result extracted from the vegetation.
5. The method of extracting ecological patches based on high-resolution images according to claim 4, wherein the vegetation extraction comprises:
taking cultivated land, buildings and water surface as masks and terrain lines as constraint conditions, and carrying out multi-scale segmentation on the remaining image range;
on the basis of the segmentation result, selecting forests, shrubs and grasslands as classification samples, and distinguishing all segmentation pattern spots by adopting a random forest method to obtain the forests, the grasslands and the shrubs; and
and performing edge simplification and smoothing on the extraction result to obtain a final vegetation extraction vector result.
6. The method for extracting ecological patches based on high-resolution images as claimed in claim 1, wherein the step S4 comprises:
merging the same type of patches of each subtask area to obtain an extracted result vector of each ecosystem type of the complete area;
updating the extracted ecological pattern spots of forest-bush-grassland-water surface-building-cultivated land on the same vector file according to an updating sequence from large scale to small scale, and then updating roads and water systems in the vector file to form an ecological pattern spot vector of all elements; and
and carrying out single-component to multi-component operation on the updated image layer to eliminate fine broken spots and long and narrow pattern spots in the updated image layer so as to obtain a final ecosystem image.
7. The method for extracting ecological map spots based on high-resolution images as claimed in claim 4, wherein the inputting of the samples into the corresponding convolutional neural network for training according to the types of the samples to obtain the corresponding farmland, building and water surface extraction models comprises: and inputting a farmland sample into the RCF network for training, inputting a building sample into the D-LinkNet network for training, and inputting a water surface sample into the U-Net network for training to obtain corresponding farmland, building and water surface extraction models.
8. The method for extracting ecological map spots based on high-resolution images according to claim 7, wherein the arable land samples are input into an RCF network for training, and the method comprises the following steps:
performing overall analysis on the global farmland plots, uniformly selecting farmland samples in an image range, drawing farmland boundary information in the farmland samples, and converting the farmland boundary information into binary results serving as label data serving as training data of an edge monitoring network;
through training boundary characteristics, obtaining an extraction model of farmland boundary information, predicting the cut images of each task area, and obtaining a boundary prediction strength result graph of each task area;
obtaining a binary image of a farmland prediction result through skeleton line extraction and broken line connection; and
and obtaining a cultivated land extraction vector result through post-processing such as grid vectorization, vector straightening, smoothing and the like.
9. The method for extracting ecological patches based on high-resolution images as claimed in claim 7, wherein the building samples are inputted into a D-LinkNet network for training, comprising:
analyzing the residence places in each subtask area, taking the residence places with similar characteristics as the same type of extraction object, making building samples with different image characteristics, drawing building areas in the samples and distinguishing the building areas from the background by using the attribute values;
converting the drawn building area sample into binary label data according to the attribute value;
inputting the building binary label data into a D-LinkNet network for training to obtain an extraction model of a residential area; predicting the image of the subarea by using the network to obtain a residence prediction intensity map of the corresponding subarea; and
and obtaining a final vector extraction result of the building area through grid vectorization and vector post-processing operation.
10. The method for extracting ecological patches based on high-resolution images as claimed in claim 7, wherein the water surface samples are inputted into a U-Net network for training, comprising:
analyzing the characteristics of the river water surface, the lake water surface and the pond water surface in the image, selecting representative rivers, lakes and ponds as samples, drawing the water surfaces in the samples, and marking the differences from the background according to the attributes;
converting the drawn sample into a binary icon label file according to the attributes, inputting the binary icon label file into a U-Net network, and training the characteristics to obtain a water surface extraction model;
predicting the image of the subarea by using the water surface extraction model to obtain a water surface prediction intensity graph of the image; and
and obtaining a final water surface vector extraction result by adopting a grid vectorization and vector post-processing tool.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210880187.3A CN115187866A (en) | 2022-07-25 | 2022-07-25 | Ecological pattern spot extraction method based on high-resolution image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210880187.3A CN115187866A (en) | 2022-07-25 | 2022-07-25 | Ecological pattern spot extraction method based on high-resolution image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115187866A true CN115187866A (en) | 2022-10-14 |
Family
ID=83521235
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210880187.3A Pending CN115187866A (en) | 2022-07-25 | 2022-07-25 | Ecological pattern spot extraction method based on high-resolution image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115187866A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116310366A (en) * | 2023-02-21 | 2023-06-23 | 中国科学院地理科学与资源研究所 | Automatic extraction method for mountain forest lines |
| CN116824157A (en) * | 2023-06-30 | 2023-09-29 | 中国科学院空天信息创新研究院 | Sampling point determination method, remote sensing product authenticity verification method, device and electronic equipment |
| CN117079157A (en) * | 2023-08-30 | 2023-11-17 | 杭州中科品智科技有限公司 | Mountain area photovoltaic panel monomer extraction method |
-
2022
- 2022-07-25 CN CN202210880187.3A patent/CN115187866A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116310366A (en) * | 2023-02-21 | 2023-06-23 | 中国科学院地理科学与资源研究所 | Automatic extraction method for mountain forest lines |
| CN116824157A (en) * | 2023-06-30 | 2023-09-29 | 中国科学院空天信息创新研究院 | Sampling point determination method, remote sensing product authenticity verification method, device and electronic equipment |
| CN116824157B (en) * | 2023-06-30 | 2024-02-02 | 中国科学院空天信息创新研究院 | Sampling point determination method, remote sensing product authenticity verification method, device and electronic equipment |
| CN117079157A (en) * | 2023-08-30 | 2023-11-17 | 杭州中科品智科技有限公司 | Mountain area photovoltaic panel monomer extraction method |
| CN117079157B (en) * | 2023-08-30 | 2024-06-07 | 杭州中科品智科技有限公司 | Mountain area photovoltaic panel monomer extraction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Voltersen et al. | Object-based land cover mapping and comprehensive feature calculation for an automated derivation of urban structure types at block level | |
| Morton et al. | Final report for LCM2007-the new UK land cover map. Countryside survey technical report no 11/07 | |
| Gamanya et al. | Object-oriented change detection for the city of Harare, Zimbabwe | |
| Ottinger et al. | Monitoring land cover dynamics in the Yellow River Delta from 1995 to 2010 based on Landsat 5 TM | |
| Mallinis et al. | Object-based classification using Quickbird imagery for delineating forest vegetation polygons in a Mediterranean test site | |
| CN115187866A (en) | Ecological pattern spot extraction method based on high-resolution image | |
| Mandal et al. | Land suitability assessment for potential surface irrigation of river catchment for irrigation development in Kansai watershed, Purulia, West Bengal, India | |
| Feizizadeh et al. | An object based image analysis applied for volcanic and glacial landforms mapping in Sahand Mountain, Iran | |
| CN107392133B (en) | Desert plant remote sensing identification method using object-oriented multi-source information fusion | |
| CN111028255A (en) | Farmland area pre-screening method and device based on prior information and deep learning | |
| KR102316598B1 (en) | Fabricating system and method of Landcover map, and Program recording media | |
| CN117475314B (en) | Geological disaster hidden danger three-dimensional identification method, system and medium | |
| Dinis et al. | Hierarchical object-based classification of dense urban areas by integrating high spatial resolution satellite images and lidar elevation data | |
| Amin et al. | Developing spatial model to assess agro-ecological zones for sustainable agriculture development in MENA region: Case study Northern Western Coast, Egypt | |
| WO2024020744A1 (en) | High-resolution image-based ecological patch extraction method | |
| Kumar et al. | Constrained linear spectral unmixing technique for regional land cover mapping using MODIS data | |
| CN116486282A (en) | Digital elevation model manufacturing method and system based on deep learning, electronic equipment and storage medium | |
| Tiwari et al. | An integrated approach of remote sensing and GIS for land use and land cover change detection: a case study of banjar river watershed of Madhya Pradesh, India | |
| de Castro Tayer et al. | Assessment of a semi-automatic spatial analysis method to identify and map sinkholes in the Carste Lagoa Santa environmental protection unit, Brazil | |
| Nagel et al. | High spatial-resolution land cover classification and wetland mapping over large areas using integrated geospatial technologies | |
| Gaber et al. | Mapping the surface changes in the area of East Port-Said, Egypt using multi-temporal and multi-sensors remote sensing data | |
| Abdulrazzaq | Used satellite images and GIS to study the degradation lands in the western desert upper Euphrates, Anbar Province, Iraq | |
| Matkar et al. | Land use/land cover changes patterns using geospatial techniques-Satara district, Maharastra, India: a case study | |
| Anwar | Land use change dynamics: A dynamic spatial simulation | |
| Driese et al. | Rule‐based Integration of Remotely‐sensed Data and GIS for Land Cover Mapping in NE Costa Rica |
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 |